Michelle Brown

Jen Yang-Wong

Anna-Sofia Lesiv

Eric Tarczynski

James Hennessy

Kyle Harrison

Jessica Retrum

Contrary HQ

Jason Chen

Will Robbins

The ability to perceive objects, discern their dimensions, and contextualize them alongside our relative position in space is a core ingredient in humans’ evolved ability to think, learn, and direct our own actions. For this reason, teaching perception to machines is key to the advancement of both artificial intelligence and robotics.

As I wrote in a past essay, robots can be thought of as computers with a body; a kind of embodied computation. Mechanical bodies have three areas of competency: sensing (computer vision), thinking (AI), and acting (robots themselves). Having covered the Acceleration of Artificial Intelligence and the Rise of Robots in previous posts, it’s now time to turn to the first part of this triad by exploring lidar, which lies on the frontier of computer vision.  

Lidar, short for “light detection and ranging” (similar to “radar”, which is an acronym for “radio detection and ranging”), has emerged as a technology that significantly improves perception for both humans and machines. While traditional visual perception relies on capturing light to create 2D images and then infusing them with depth information, lidar captures a 3D model of its environment from the get-go. It does this by shooting tiny laser beams and measuring the time it takes them to reflect off of surfaces to calculate the distance that the laser beams traveled. By shooting hundreds of thousands of little laser beams at a time, lidar can create a fully accurate 3D scan of its surroundings.

Though its core technology has been around for over half a century, it’s only within the previous twenty years that lidar has found its killer app: autonomous driving. The demand from the self-driving industry has transformed lidar from a rarity to an increasingly ubiquitous technology that’s already installed in all modern iPhones.

It unlocked new vistas, allowing us to do everything from measure continental drift, dock spacecraft to the ISS, and discover lost cities. But there’s still a lot of potential for further advancement with lidar, because the economics of scaling haven’t even fully kicked in yet. With the prospect of Moore’s Law-like improvements in precision and cost for lidar technology over the coming years, it’s high time to review the history of lidar and explore its potential impact across the future of industries and consumer goods.

At the core of lidar technology is the ability to control the direction of light. Historically, light mystified philosophers and scientists. It wasn’t at all clear where light actually came from. The Sun produced it, as did fire; but how? Strikes of lightning and electricity seemed to create it too; but why?

It would take until the First World War for these questions to finally be answered by the independent efforts of Niels Bohr and Albert Einstein. Bohr had been studying the structure of the atom and was conducting spectroscopy experiments where he would take gasses like hydrogen or nitrogen gas and expose them to an electrical current. When the gas started to glow, he would direct the light through a prism to reveal the discrete colors emitted by the gas.

These gasses kept producing the same visible colors. This led Bohr to conclude that atoms must contain electrons at varying discrete energy levels which correspond to specific wavelengths of radiation.

In other words, Bohr noticed that after an atom’s electrons absorbed energy, they released that energy in the form of electromagnetic radiation, or light. The exact wavelength of light they released depended on the energy level of the electrons.

A few years later, building off of Bohr’s work in 1917, Albert Einstein developed a theory called “spontaneous emission” which described exactly how an atom’s electrons produce light.

According to Einstein, electrons absorb energy from photons to get to higher energy states. When the electrons return to their original ground energy state, they release the initial photon in some random direction as electromagnetic radiation or light. The wavelength of the photon released is equal to the energy differential between the higher energy state and the ground state.

In that same paper, Einstein also wondered what might happen if you kept bombarding an electron in an elevated energy state with more photons. Einstein predicted that if you kept shooting photons at an already excited electron, the photons it would emit would not radiate out randomly, but rather move in the same direction as the photons shot at it. Furthermore, the rate of overall photon emission would be amplified due to its already excited state! The theoretical result would be an amplified light beam traveling in the same direction. He called this process “stimulated emission.”

Spontaneous emission describes the orb-like glow we normally experience with visible light as it shoots out in random directions, but stimulated emission imagined what it might take to get light waves to move neatly in the same direction.

What’s remarkable is that Einstein predicted this behavior from a classical mechanics perspective. The real reason for photons moving in the same direction under stimulated emission is actually explained today by the very complicated laws and mathematics in quantum field theory. How exactly, lacking the tools of quantum mechanics, Einstein was able to accurately describe the supremely counterintuitive behavior of stimulated photon emission is beyond us. In any case, his prescient ideas gave us the clues to how we might one day be able to control light for our own devices.

Designing a device to test out Einstein’s ideas of stimulated emission would take almost four more decades. In 1953, Charles Townes, James P. Gordon, and Herbert J. Zeiger were the first to validate the theory of stimulated emission by building a device called a “maser,” short for “microwave amplification by stimulated emission of radiation” which excited a column of ammonia gas to emit microwaves in a parallel direction. 

There was a great deal of interest in learning to direct electromagnetic waves at specific targets after the invention of radars during World War II, which were deployed to great effect on the battlefield. Radar designs at the time were simple and effective. A basic radio antenna produced a stream of radio waves radiating outward. When the waves hit a reflective object, they would reverberate back in the direction of the antenna, and eventually be picked up by a receiver. The receiver measured how much time it took for them to travel from the reflective surface, revealing the distance to the object. It turned out that radio waves could bounce off of metal surfaces, which proved rather useful when trying to do covert reconnaissance or detecting the presence of enemy vehicles on a battlefield. 

The problem with radio waves was their giant size. Radio wavelengths could span kilometers, which didn’t provide much precision as to the precise coordinates of an object detected by radar. As a result, there was a mad post-war dash to fund research on electro-optics and range detectors which used smaller wavelengths to detect objects with greater accuracy. The invention of the maser eventually won Townes the Nobel Prize in Physics, but it was just a stepping stone to an invention that would truly change the world forever — the laser.

Townes and his colleagues tried for years to develop a functioning laser through stimulated emission, but it was ultimately a Stanford physics graduate named Theodore Maiman, working his first job at Hughes Aerospace in Culver City, who figured out the trick in 1961. 

Maiman, who had been studying the use of crystals in optics, got the idea to use a synthetic ruby crystal and coil it with a flash tube that would excite the electrons in the ruby crystal. The entire contraption was book-ended by mirrors, one of which held a tiny opening. When Maiman shot a stream of photons through the device, a concentrated beam of ruby red light came out the other end. Thus; the laser was born.

Immediately after Maiman invented the laser, his employer, Hughes Aerospace, began investing in commercializing the technology. Its first product was an early lidar-like tool called the “Colidar,” built in 1963, which stood for “coherent light detection and ranging.” It operated like a highly precise range-finder. In military scenarios, these were used by combatants to measure distance to objects of interest to determine if they were within artillery range. The Colidar was capable of shooting a beam of light up to 7 miles out, and had a receiver able to detect objects with an accuracy of up to 15 feet.

All through the sixties, laser technology was improving. The sixties was a decade of incredible progress in electronics, leading engineers to experiment with producing all kinds of new lasers. Gas lasers and even semiconductor lasers quickly became a reality, eventually leading the way new applications including spectroscopy, laser surgery, and even industrial applications like cutting and welding with light.

As lasers got more powerful and peacetime set in, the public began seeing the utility of laser sensing and detection in other contexts. The first lunar laser ranging experiments were conducted at MIT in 1962, which shot laser beams at the moon to determine its distance from the Earth. By the 70s, laser altimeters were standard in the toolkits taken by astronauts to the moon across the Apollo missions.

The development of faster-pulsing lasers throughout the nineties made it possible to use  laser light to sweep across an entire horizon, rather than just targeting one specific point. By rotating the lasers through mechanical motion, lidar systems could detect objects of all sizes across a full horizon. When mounted on planes, lidar systems were able to produce some of the first topographical maps of the Earth’s surface. In 1996, the Mars Orbiter Laser Altimeter, or MOLA, was sent to orbit around Mars, and after a period of four and a half years, succeeded in completing a full 3D scan of the planet. 

Though these were impressive achievements, lidar systems were few and far between. Most were housed in universities or commissioned by massive government agencies. The appeal of lidar was initially its military applications. However, with the influx of GPS and other kinds of communications technologies, lidar lost its killer use-case. The technology then stagnated and was largely forgotten about until about 2004, when things changed once again.

In 2004, DARPA, the Defense Department’s Research arm, hosted its first “Grand Challenge,” a competition meant to encourage the development of autonomous vehicles. The US Congress authorized the challenge with the hope that by 2015, a third of the army’s ground forces would be replaced by autonomous machines. DARPA structured the Grand Challenge as a race. The first self-driving car to complete a 150 mile route in the Mojave Desert would win $1 million. However, the outcome was very disappointing. None of the teams finished. The best performance was by Carnegie Mellon’s autonomous car, which managed to get 11 miles before hitting a rock and stopping. No prize money was awarded. 

The biggest problem with self-driving was getting the car to perceive the route and obstacles along it. Computer vision was a difficult problem AI researchers across the country were struggling to solve. Computers could begin to recognize discrete objects from an image, but discerning depth of field from a two-dimensional canvas was a really challenging problem. It seemed that nothing less than a fully-functioning brain would be capable of sophisticated enough perception for autonomous vehicles. With only a few months to go before the 2005 Grand Challenge, the teams figured that something needed to change. Rather than using cameras for vision, the Stanford team decided to place a lidar detector on top of its car. 

Cameras see color, but lidar sees distance. With a lidar system, computers could see in all three dimensions. . It was a brilliant idea with a big reward at the end. That year, five cars completed the track, but it was Stanford’s lidar-enabled car “Stanley” that got the fastest time — six hours and fifty four minutes. Stanford’s team won a prize of $2 million, and the advantage of lidar for self-driving was cemented in history forevermore. 

Since then, numerous self-driving startups have spawned, jumping at the opportunity to accelerate the capability with the brilliant help of lidar technology. The only problem was that there was no off-the-shelf lidar product they  could buy to get started. Most lidar systems were terribly expensive and clunky. To create a 360-degree field of view, a laser emitter and receiver had to be  mounted on top of the car, and rotated mechanically.

Velodyne was one of the main manufacturers of such systems, which they sold for $75,000 a piece prior to 2017. Major self-driving operations from Alphabet’s Waymo to Caterpillar and Ford Motors outfitted their cars with Velodyne’s five figure spinning lidar systems. These were complete with 64 spinning lasers and capable of generating hundreds of thousands of data points per second at a range of up to 200 meters.

The early days of self-driving were odd in that way. The sensors cost more than the cars themselves, and just because the lidar could produce a three-dimensional “point cloud” representing a field of vision didn’t mean that there was mature software capable of classifying the objects in the line of sight. The situation led to Elon Musk making his now-famous statement that “anyone relying on LIDAR is doomed.”

As a result, Tesla decided to go in another direction than lidar for its own autonomous efforts. In 2014, Tesla retrofitted its Model S cars with 360-degree cameras as part of its new advanced driver-assisted features, which had the additional benefit of logging driving data to continue training its camera-based self-driving model.

After this, the lidar battle lines were drawn. On the other side of the schism, teams were doubling down on lidar. By 2017, after seven years of tinkering, Alphabet’s Waymo had begun producing its own lidar systems in-house, and achieved remarkable cost savings. Its home-made systems ended up costing only $7,500 to produce, a 10x improvement from just a few years ago (but still more expensive than Tesla’s cameras).

Though the exact design of Waymo’s sensors is not fully known, Alphabet’s team likely had many novel options to choose from as a range of cheaper, and more energy efficient lidar systems had emerged following Stanley’s success in the 2005 Grand Challenge.

In particular, three approaches towered over Velodyne’s approach to lidar:

The microelectromechanical mirrors (MEMs) approach: Rather than spinning lasers around, this approach fires the laser beam onto a tiny mirror that spins, allowing the laser to capture the full horizontal field of view. This is essentially a miniaturized version of the classic lidar system. Another tiny mirror can be added to redirect the laser beam along a vertical orientation, too, but once more moving parts are added, the complexity of the system begins to grow. Smaller size and cost are convenient here, but having any kind of moving parts inevitably exposes your system to wear-and-tear. Mirrors are also prone to shock and vibrations from natural car movement, which could hinder performance.


Flash Lidar: Rather than emitting hundreds of thousands to millions of tiny beams of laser light in all directions, flash lidar floods the entire field of view with one large laser pulse. It yields a more compact system but requires a pretty powerful burst of light. The power of the light is limited by the fact that this light can’t be blinding to human eyes, which unfortunately limits the detection range on these systems. Currently, the detectors for eye-safe flash lidar are still rather pricey, which is another hiccup with this option. At the moment, flash lidar systems are only truly appropriate on larger systems like planes conducting aerial surveys.


Optical Phased Arrays: A full-lidar system with no moving parts and no soldering required is one of the most exciting and promising new system designs on the market. This is the future promised by optical phased arrays. Based on the calibration of phase shifters, which are controlled electronically, a single laser beam can be quickly directed to scan a wide field of view. The beams can be steered on a horizontal and vertical axis all without any moving parts. To do this, millions of array elements are required, which need to be placed microns apart. Although this sounds rather challenging, with contemporary photolithography techniques developed by the microchip industry, the technology to do this has already been mastered. For this reason, optical phased arrays printed on silicon wafers  are being called “optics on a chips.” Just as microchips had their era of Moore’s Law scaling, many foresee the same happening for these miniaturized lidar systems. As the scale of manufacturing increases, the costs of these chips and lidar systems will decrease dramatically.

Lidar technology has emerged from the shadows of obscurity to claim its well-deserved moment in the spotlight. In his book “The Laser That’s Changing the World,” Todd Neff calls Lidar a “most striking macroscopic application of quantum physics.” The ability to laser scan our physical surroundings have already enabled us to do extraordinary things. 

Powerful, airborne lidar systems can detect sea levels and melting glaciers, and can produce scans of entire planets. Such systems give us the ability to assess the terrain of the Earth from above even under dense tree cover. One recent application of this has been uncovering lost ancient cities, like one discovered in May 2022 hidden deep in the Amazon forest, imperceptible to satellites or even passersby, but incapable of fooling the eyes of lidar drones.

Lidar is used to measure distance in space. Today, it already assists SpaceX’s Dragon capsule to navigate and dock with the International Space Station. One day, its applications in space might be even more expansive. One startup called Nuview, for example, is interested in launching a fleet of satellite-borne lidar systems to create a full 3D scan of planet Earth, akin to the model we already have of Mars. 

Many might also be surprised to hear that lidar is already installed on most of our phones. All iPhones past the model 12 have been installed with a miniaturized lidar system capable of creating 3D scans at a range of up to 5 meters. Smaller range lidar capabilities are easier problems to solve than larger systems that need to negotiate large ranges, quick point processing, all with eye-safe laser beams. Sony is Apple’s present lidar system vendor, and is already mass manufacturing a suite of small lidar systems that can be added to any phone or device.

Lidar is one of those technologies that we have all the components and technical know-how to produce, but delayed doing so at scale because the economic incentive to do so was missing. Over the previous decade, the pursuit of autonomous driving has acted as a catalyst for the rapid advancement of lidar technology, and since then lidar development has unleashed a wave of innovation, reminiscent of Moore's Law scaling effects, with continuous improvements in sensor accuracy, range, and affordability.

Looking to the future, lidar has the potential to become an indispensable feature in wearable devices for professions such as police officers and firefighters. Equipping professionals with wearable lidar technology can enhance their situational awareness, enabling them to navigate complex and hazardous environments with greater precision and safety.

Lidar may also extend to the realm of motion capture, where the technology's ability to capture fluid, real-time 3D motion has the potential to massively decrease costs in the entertainment industry. Instead of relying on cumbersome and time-consuming methods such as actors wearing motion-capture “mocap” suits, lidar can seamlessly record and digitize human movements, leading to more efficient and realistic CGI-enhanced videos and animations. 

Real-time monitoring can also lead to new applications in construction and engineering. The ability to capture and analyze construction progress in real time would mean better project management, on-the-fly adjustments, and a more comprehensive understanding of structural integrity. In the fields of manufacturing, lidar's fast 3D scanning capabilities can significantly reduce manufacturing times by facilitating quick and precise scans of components or component models, letting you go from 3D scanning to 3D printing within a few minutes.

And of course, lidar is already playing an important role in enhancing robotic perception. With the integration of lidar sensors, robots can swiftly and accurately recognize and understand their environment in three dimensions, enabling them to operate autonomously and navigate complex surroundings with greater efficiency.

As lidar technology continues to advance and find its footing in an array of applications, it is evident that its transformative potential reaches far beyond its original conception. The convergence of lidar with various disciplines promises a future of precision, efficiency, and discovery.

Below are the companies building at the frontiers of lidar technology:

1. Ouster sells off-the-shelf lidar systems and is working on mass producing solid-state lidar chips. In early 2023, Ouster acquired industry veteran Velodyne.

2. Luminar is a leader in building solid-state lidar and software perception systems.

3. Blickfield builds various types of solutions from solid-state to MEMs-enabled lidar hardware and software.

4. Neuvition works on developing solid-state lidar systems alongside selling off-the-shelf lidar solutions.

5. Nuview hopes to outfit satellites with lidar systems to conduct earth observation and monitoring for industrial use.

6. Aurora recently purchased Uber’s autonomous vehicle branch, and is working on lidar-enabled autonomous passenger and freight vehicles.

7. Cruise is working on developing a fleet of lidar, radar and camera-enabled self-driving passenger vehicles.

8. Waymo develops its own lidar sensors in-house, amid an effort to one day produce a fleet of lidar-enabled self-driving robo taxis.

A deep dive into a lidar; a once-forgotten technology that's now transforming the field of computer vision.

Lidar and the Future of Computer Vision

Ever since Isaac Asimov coined the term “robotics”, we’ve been bracing ourselves for a world where robots are everywhere. The Jetsons had us expecting Rosie the Robot would be taking care of all our household chores. The reality turned out to be starkly different.

Rather than robots living among us, the vast majority of robots we do have are mostly hidden from view. They are sheltered in factories where they toil away screwing caps onto bottles or sorting components.

Though automation is increasing across every aspect of life, we don’t yet trust robots to act fully outside of our supervision. In the most crucial situations, robots act as extensions of ourselves. Remote-controlled drones, interplanetary rovers, disaster relief bots, and even surgical robots allow us to operate in dangerous or critical environments with more accuracy. However, even the cute delivery robots that scoot alongside us on sidewalks are, more often than not, human-operated.

The greatest challenge within the field of robotics thus far has been getting robots to think for themselves and make decisions in dynamic environments where they have to deal with uncertainty. Though progress on this front has stalled for decades, recent breakthroughs in AI driven by success with large language models are re-invigorating research in novel robot training approaches, signaling a possible new spring in robotics research.

The main thing distinguishing a computer from a robot is the fact that robots can act upon the physical world. Robots are computers with a body, a kind of embodied computation. 

To interact with the world, all robots must have three main competencies; sensing, thinking, and acting. As a result, the field of robotics itself can actually be subdivided across these three main buckets.

Sensing lies on a spectrum of complexity. The simpler side of this spectrum involves tactile mechanisms like switches that go off when they bump into an object or proximity sensors that detect the presence or absence of an object. These cheap and light mechanisms are widely used in smaller robot systems like the Roomba, for example.

However, for decades, researchers have striven to give robots more sophisticated methods of perception. For instance, the field of artificial intelligence began with efforts to get computers to develop visual perception and recognize images. For a long time, such  systems of computer vision were only capable of “seeing” in two dimensions. However, with the development of light detection and ranging (LIDAR) and radio detection and ranging (RADAR) in the early 2000s, top-of-the-line systems are now capable of seeing in 3D just like us.

While vision has been the most complicated sensory faculty to mimic in computers, everything from sound to touch – and now even smell – are also being actively explored.

Thinking is one of the outstanding faculties that we have yet to get right. The field of robotics has had tremendous success with the physical engineering side of robotics, particularly in cases where the robots were hard-coded to perform repetitive, highly predictable tasks. Teaching robots to behave appropriately in dynamic and uncertain environments, on the other hand,  has always been a very difficult undertaking. Part of the challenge can be chalked up to the fact that physical activity is very complicated. Though most of us don’t consciously think about it, standing on our own two feet involves a careful balancing act, shifting around our weight in line with the signals from our vestibular system which orients our body in space. Gary Zabel, a lecturer at UMass, writes, “It requires more neurological resources in terms of sheer brain mass to walk on two legs than it does to play chess or prove mathematical theorems, both of which computers have been doing since the 1950s.”

If you want a visual guide to the trials and tribulations of getting physical behavior right, just take a look at this compilation of robots losing their battles against gravity.

Actuators create the physical form factors of robots. Their role is to convert energy into mechanical force. 

The design of actuators determines what robots can do and how they do it. Actuation is also where the challenges that set  robotics apart from traditional computer science become immediately apparent. As soon as you settle on a physical form, everything that your robot can do must conform to the constraints of its actuator design. This is why it’s more straightforward to design arms for robots that do well-specified and repetitive tasks, versus arms for robots expected to operate in a general fashion.

The fundamental components of actuation are the engineering primitives that combine to produce a machine capable of the desired movements. When designing a robotic arm or mechanism, engineers can choose between using electric motors (useful for smaller, less energy-intensive applications), hydraulic actuators (which use pressurized fluid to produce a great amount of force and are often used in industrial settings), or pneumatic actuators (which use compressed air to produce movements required for quick, precise activities). 

Design of a robot’s physical components, and the nature of the materials they use, are often biologically inspired. Take for instance the Gecko robot that can scale windows, or the robot arm that elongates and tenses up just like a human muscle. 

Material design also plays a big role in creating new robotic primitives. Some examples include the research being conducted on shape memory alloys or piezoelectric actuators, which are both materials that can change shape when exposed to heat or electrical current, and can allow for incredibly precise movements.

The combined abilities to sense, think (however primitively), and actuate takes you most of the way toward making any kind of robot you want. Robots already operate in the air, in space, and underwater. They are used in everything from disaster response to entertainment and even medicine. Though there are many specific kinds of possible robots, the two biggest and most influential categories are industrial robots and autonomous mobile robots.

Industrial robots are designed to perform a range of tasks that are too difficult, dangerous, or repetitive for humans like lifting extreme weights, working in extreme temperatures, or manufacturing with extreme precision. However, such robots are typically built for operation in highly-structured environments with very few unknowns.

In order to be produced more easily off-the-shelf, industrial robot form factors usually come in one of four major designs; articulated arms, parallel arms, cartesian arms, and selective compliance assembly robot arms (SCARA arms). Each of these can then be configured to perform the exact task you want them to — albeit at quite a cost.

Not only is industrial robot configuration expensive, it’s rather involved. There are robot parts suppliers like ABB, KUKA, Toshiba, EPSON and so on, but these companies just specialize on the basic hardware. The robotic arms they produce are like a drill with no drill bit. The drill bit determines what the drill could ultimately be used for.

With industrial robotics, the drill bit part is called the “end effector,” and these are typically designed to handle a specific  task. However, finding an end effector is not the end of the story. There’s all kinds of other set ups that industrial robots need. Basically, the entire factory usually ends up being designed around the industrial robots it uses.

As Dmitriy Slepov, co-founder of Tibbo Technology, put it: “Robots can’t (yet) run to the warehouse and cart back a bunch of parts. Your robot is like a master craftsman sitting in the middle of a studio. Everything must be brought to it. For small parts, such as screws, you will need to install “screw presenters” — machines that “offer” screws in the right orientation. Larger parts will need to come on conveyor belts or some other means of transportation”.

Robots often need to be retrofitted with additional sensing systems, and must also have adequate guardrails set up around them so as to not endanger any humans. That’s a lot of steps involved in achieving factory automation, and a lot of expertise is required to accomplish this. That’s why there’s a market that exists just for configuring industrial robots to factories. It’s an industry called system integration, with big players like CHL Systems which specializes in manufacturing systems for goods and FANUC North America which specializes in designing factory floors for car manufacturing.

After all that, you’re still left with the hard truth that none of these robots can think for themselves. All of their actions and abilities will need to be hard coded into them. This means that all the work required to set up an automated manufacturing process has to be thrown out the window in the event that a change to any part of the process is needed. An entirely new system will need to be designed when this happens. Robots’ inability to switch between tasks easily makes massive investments in automated systems like these quite risky. 

Unlike robots, humans can switch tasks pretty easily and adapt to new constraints and environments almost instantaneously. In a dynamic factory setting, this is the reason that human labor is currently still much more effective than robot labor – it is more easily and more cheaply reconfigurable. One way to take advantage of human nimbleness and robotic precision and power is through collaborative control, in other words, using robots that are partly operated by humans. Unsurprisingly, collaborative robots are the most commonly ordered robots in industrial settings.


As exciting as industrial robots are, if you’ve ever seen Boston Dynamics’ robot dog, Spot, climb a flight of stairs, or watched their bipedal robot, Atlas, do parkour, you know that this is where modern robotics gets entertaining. The holy grail of robotics research is to get robots to move and think on their own.

Robots like Spot and Atlas can move on their own. In fact, these specimens have gotten very good at basic locomotion and balance whether on four legs or on two. Yet though they’ve been designed to traverse varied kinds of terrain and perform many sorts of complex movements, they still struggle with navigating in unfamiliar environments. 

Another major branch of autonomous robotics is in self-driving cars. This technology has shown tremendous progress over the past few years, particularly because of the number of cars on the roads capable of recording driving behavior to train autonomous models the elements of good driving. However, determining what ‘good driving’ really means itself remains an unresolved challenge. For instance, is an autonomous vehicle that’s as good as a human driver enough, or would we want an autonomous vehicle that is substantially better than a human? From this perspective, we’re still likely a ways away from having fully autonomous self-driving cars on our roads. Even other types of seemingly autonomous robots, like Mars rovers or even cute sidewalk delivery bots, are, for the most part, still human-operated.

The history of robotics is deeply rooted in the history of industrialization. The Unimate, for instance, was the first ever industrial robot. It was developed by George Devol and Joseph Engelberger specifically for General Motors in the fifties, and within the next couple of years was featured on assembly line floors, doing tasks like spot welding or handling hot metal parts in car manufacturing.

The large industry budgets and iterative cycles from manufacturing applications gave the field of robotics a terrific canvas on which it could apply many of its theoretical ideas.

Many of these ideas came from a branch of mathematics developed in the thirties called control theory. Control theory was concerned with mathematically describing the feedback mechanisms and optimization techniques which contributed to equilibrated systems. The ideas of control theory were instrumental in designing the mechanisms by which robots incorporate information from sensors to maintain stability and adapt to changes in environment.

The development of better physical primitives, like electric sensors and actuators, meant that this information could be communicated faster and more efficiently, in contrast to older and more rudimentary robotic systems which relies on hydraulic or pneumatic actuation. By the 1980s, robots had already become commonplace in factories, performing jobs like assembly, welding, and inspection, but it was not until rather recently that the proliferation of robots really took off.

The graph of annual shipments of robotics hardware paints a picture of an industry that was seeing rather sluggish growth until just about a decade ago, when global labor prices began to increase while the cost of manufacturing electronics kept falling. In 2019 roughly $40 billion was spent globally on purchasing and installing industrial robotics equipment. All signs point to this only continuing to grow. 

The big outstanding challenge for robotics is teaching machines how to make independent decisions about physical movement. To date, most artificially intelligent systems have been trained entirely on text. As a result, those systems tend to be ill-prepared to deal with the physical inputs of the real world.

For instance, though large language models like GPT-3 are very good at writing cogent poems and essays, it is clear that they don’t know how the things they write about translate to the physical world. To demonstrate its shortcomings, two researchers asked ChatGPT a series of questions testing its physical reasoning abilities: 

“When we asked GPT-3, an extremely powerful and popular artificial intelligence language system, whether you’d be more likely to use a paper map or a stone to fan life into coals for a barbecue, it preferred the stone. To smooth your wrinkled skirt, would you grab a warm thermos or a hairpin? GPT-3 suggested the hairpin. And if you need to cover your hair for work in a fast-food restaurant, which would work better, a paper sandwich wrapper or a hamburger bun? GPT-3 went for the bun.”

A number of roboticists believe the reason for this is that embodiedness is a precondition for developing general intelligence. For instance, Boyuan Chen, a robotics professor at Duke University, has said “I believe that intelligence can’t be born without having the perspective of physical embodiments.” 

However, there’s still a powerful camp that believes large language models haven’t yet reached their full potential. Though GPT-3 failed at spatial reasoning, GPT-4 answered all the same questions correctly. It knew that paper was better at fanning coals than a stone was and that a sandwich wrapper works better as a hat than a hamburger bun. Notably, GPT-4 was trained on a combination of text and images, which suggests that language models could still get a lot better if only given a little more context about the text they’re manipulating. These two perspectives outline the two main approaches to teaching robots how to behave in unstructured environments; hierarchical learning, and language-based learning.

Hierarchical learning builds off of the intuition that any complex task, like washing the dishes, is composed of a series of simpler tasks organized in a hierarchical order. Washing the dishes involves picking up a dish, scrubbing it with soap, rinsing it with water and then drying it in precisely that order. Each of those sub-steps is itself composed of smaller tasks in turn. Picking up a dish, for example, might involve identifying the dish, moving toward the dish, and using the appropriate arm movements to pick it up effectively without dropping it.

Stefan Schaal at the University of Southern California has developed one of the best known frameworks to teach robots motor skills through hierarchical learning. Known as the dynamic movement primitives approach, it amounts to showing robots a number of primitive movements, which when combined with reinforcing techniques, eventually teaches robots to combine the movements together to achieve more complex goals. 

One of the biggest problems with this approach, however, is that it’s inefficient. Artificially intelligent systems require lots of examples in order to learn, which means large data sets are needed to train them. This means teaching robots using movement primitives involves a lot of time instructing them on how to perform basic movements and repeating those over and over. Given that each robot has a unique physical structure, there is limited transferability across robots with other form factors. For them, the entire learning process would need to be repeated.

The time demands of this learning approach, so far, have resulted in the ultimate failure of many of the commercial programs that adopted it. Notably, OpenAI once had a robotics team that achieved the notable success of training a robot hand to solve a Rubik’s cube. However, given the limited transferability of this knowledge, and the more rapid development of large language models, OpenAI shuttered its robotics team in 2021.

Language-based learning, on the other hand, begins with the insight that complex tasks could just as easily be broken down into simpler instructions by a sophisticated language model. In fact, breaking down the task of washing the dishes is an assignment that even GPT-3 could do very well. Then, training robots on a combination of text, images, and movement primitives would ideally help the robots learn how to combine the primitives together more efficiently to accomplish tasks. Eventually, these systems learn to use discrete or continuous actions as tokens to create meaningful sequences, just as GPT-3 uses words as tokens to create meaningful sentences. 

This approach is the most novel approach to training robots and has taken off in recent years. DeepMind’s Gato is just one example of a successful implementation of this approach. This system, when attached to a number of different output sources, like a robotic arm or a video game controller, can decide whether to output joint movements, button presses, or other movements based on the context of the prompt. Taken together, the same system can play stack blocks or play Atari just as easily as it can recognize images and respond to queries.

A research team at Google, in collaboration with Everyday Robots, developed a similar system called SayCan. The SayCan system works by grounding actions expressed in language in physical pre-trained behaviors. When the language model is combined with a robot, the robot acts as the language model’s hands and eyes. Whereas physical pre-training teaches the robot how to perform basic low-level tasks, insights from the attached language model teach the system higher-level knowledge about how to combine the low-level tasks into more complex functions.

Given the historic shortcomings of robots, roboticists think it’s premature to assign anthropomorphic features to them. Joe Jones, the inventor of the Roomba, for instance, has said that “when designers festoon their robots with anthropomorphic features, they are making a promise no robot can keep.” Though this has been the prevailing view among certain roboticists for years, these days we’re increasingly recognizing that robots do have some humanoid characteristics. In particular, computers can now talk just as we do.  

Outside of the recent dash to retrofit existing robots with human-language communication, as seen with products like Moxie, a robot that uses GPT-3 to teach children, or Engineered Arts’ Ameca, which can speak and even do facial gestures, there seems to have been a rebirth in interest in building human-resembling robots. 

For one thing, it looks like OpenAI is re-entering the robotics game given its recent investment in 1X Technologies, a company developing a new bipedal robot called Neo, which should be released in Summer 2023. Figure is yet another company currently working on building a “feature-complete electromechanical humanoid" with the ultimate goal of “integrat[ing] humanoids into the labor force.” Figure’s prototype of the “first commercially viable general purpose humanoid robot” currently stands at 5 foot 6 inches, and weighs 132 lb. Figure is hoping it will be able to lift 44 lbs, walk at a speed of roughly 2.7 mph, and operate for up to five hours on a single charge.

Tesla Motors is yet another company taking a stab at building humanoid robots. On its investor day in 2023, Tesla unveiled its human-resembling robot Optimus, walking for the first time. The company hopes that Optimus, too, might one day enter the labor force. At one point, Elon Musk even said he thinks that Tesla’s robot business might one day supersede its car business. Given that Hyundai Motors is the owner of Boston Dynamics, it seems vehicle manufacturers are starting to become big robotics companies in disguise.

Given recent advances in language-based learning, robotics is once again rapidly evolving. As technology advances and costs decline, we can expect to see more types of robots becoming economically viable. With the constant expansion of the toolbox for robot designers, the possibilities for innovation are limitless. 

The challenge now is to create a robot equivalent to human labor in cost and ability, either through the development of an AGI that can figure out how to make robots work or by training robots physically to achieve AGI. As we continue to push the boundaries of what robots can do, we must also consider the ethical implications of creating machines that could potentially replace human workers. Ultimately, the future of robotics holds great promise, but it is up to us to navigate the challenges ahead with foresight and responsibility.

Below are the companies building on the frontiers of robotics:

Boston Dynamics is a subsidiary of Hyundai corporation, researching athletic mobility in robots for a variety of consumer and industrial applications.

Figure is building a humanoid robot which is hopes might one day join the labor force.

1X is building a humanoid robot which they hope will one day augment human labor.

Covariant is building turnkey artificial intelligence systems to integrate into picking and sorting systems in warehouses and factories.

Intrinsic is a subsidiary of Alphabet working on solutions to make training robots more intuitive.

Intuitive Surgical is a leader in designing and manufacturing robotic surgical systems.

Nuro is building a suite of self-driving delivery vehicles, now cleared to test on the streets of California.

Right Hand Robotics designs and builds robotic picking systems for e-commerce order fulfillment and warehouse automation.

Rapid Robotics has built a system called the Rapid Machine Operator, outfitted with advanced vision technology, which can more easily be incorporated into  existing manufacturing lines. 

A deep dive into the past and future of robotics.

The Rise of Robots

Contrary’s core mission is to support exceptional people relentlessly. Every now and then, we find companies that align with that mission very closely as well. And when we do, investing in them is a no-brainer. Replit is one of those companies.

While more than 5 billion have access to the internet, there are only about 27 million software developers—less than half a percent of the global population. As software becomes more pervasive in every aspect of people’s lives, there is a massive gap between the people who can benefit from technology, and the people who can build it.

Replit exists to “bring the next billion software creators online.” I’ve had the chance to know Amjad for several years now, and when I joined Contrary investing in Replit was my first priority. When we first invested in Replit, most people saw the product as a toy. Something to hack together side-projects on the weekends. But things have really picked up.

Just in the last year, Replit has launched Ghostwriter, an AI-enabled code assistant, a 100 Days of Code on-ramp for new developers, a mobile app, bounties for finding other programmers to help you build aspects of your product, and Replit Deployments to host your applications.

We also had the chance to cover Replit for Contrary Research. You can learn more about the company in our full memo here.

Replit has quickly built an end-to-end developer environment with all the tools you need to take an idea from development to deployment. Contrary has always been about identifying exceptional people, and now Replit has come along giving those people all the tools they could need to build their ideas.

Today, Replit has announced a $97 million round backed by Contrary, a16z, Khosla, and many more. And based on the pace the Replit team is setting, we’re just getting started!

Investing in Replit

At Contrary, we value intellectual curiosity and a broad understanding of the world around us. So it's with great pleasure that we welcome Sachin Maini to our team as an Editor to work across Contrary Research, Foundations & Frontiers, and more.

Sachin is easily one of the most well-read members of the Contrary team, having studied European History, Classics, and Literature. But he's spent the bulk of the past decade applying his talents in tech and the venture world, previously serving as a content lead at NFX and elsewhere.

In the coming months and years, we plan to continue developing nuanced and deep narratives to describe what's happening in Silicon Valley and beyond. Stay tuned for more to share here.

In the meantime, please join us in extending a warm welcome to Sachin as he begins his journey with us!

We're thrilled to welcome Sachin to the team as Editor!

Welcoming Sachin Maini

Since the advent of modern medicine, our biggest strides have come from finding effective countermeasures against pathogens like bacteria and viruses. Collectively, such pathogens had been responsible for sowing history’s deadliest illnesses from the plague, to tuberculosis, smallpox, and the flu. Ever since the invention of antibiotics and vaccines, however, we’re no longer as defenseless against these pathogenic diseases as we once were. But as humans began to live longer, new diseases took their place.

The most intractable contemporary diseases are those produced by the body itself, where either the body begins to treat its own tissue as a pathogen (auto-immune diseases like Type 1 diabetes, arthritis, multiple sclerosis, and Alzheimer’s), or where cells begin to grow uncontrollably (cancer).To date, pharmaceuticals have only proven capable of either slowing the progression of these conditions or of subduing their side effects, but have limited effectiveness against their underlying causes. To change this, an entirely new approach is needed: cell therapies. 

Cell therapies promise a total paradigm shift within medicine. Rather than relying on drugs for treatment, the idea behind cell therapies is that the cure-all for any disease can already be found within our existing cells. By genetically enhancing our cells, emerging research points to the fact that nearly any condition – from organ failure to auto-immune disease, and even cancer – might soon be a problem of the past. Cell therapies wouldn’t only transform the landscape of modern illness, but could completely change the medical industry as we know it. 

Just as there are millions of different cells in the human body, there are many types of cell therapies. However, the most well-studied and powerful types can be bucketed into two categories: stem cell therapies and immune cell therapies.

Every human being begins as a bundle of stem cells. In the womb, personhood evolves from a small group of undifferentiated cells into a complex system of specialized skin cells, nerve cells, and muscle cells which ultimately make us up. While most cells have completed their differentiation by birth, some undifferentiated stem cells remain. These stem cells can actually be found in most of the tissues that are able to naturally repair themselves like skin, bones, or even the linings of certain organs like the liver and intestines.

Harnessing the power of these cells to repair and regenerate tissue has opened up many exciting prospects in medicine. Take, for instance, heart disease. Heart disease is the leading cause of death for most men and women in America. Though surgery and treatment can help a patient survive a heart attack, damaged heart tissue can’t be regrown. Instead, rigid scar tissue grows in its place, which eventually makes the heart stiff and incapable of properly pumping blood through the body. Apart from a full heart transplant, traditional medicine, and surgery, there are limited options for patients experiencing heart failure.

However, stem cell therapies offer a promising new avenue for regenerating tissue that can’t naturally heal on its own. A number of small-scale experimental trials have demonstrated the potential of stem cells to help regrow healthy heart tissue or limit the amount of scar tissue produced after a heart attack. 

One trial in 2012 did so by extracting a patient’s own stem cells, differentiating them into heart tissue, and re-injecting them back into areas with heart damage. After six months, the patients who received this therapy showed reductions in scar mass and increases in viable heart mass. However, not all trials have yielded straightforward results. Another experimental trial concluded just one year later, showed no major improvements in patients who received bone marrow-derived stem cells injected into damaged heart tissue. Despite the conflicting results here, researchers are holding out hope — but the small scale of experimental trials makes it difficult to identify exactly what’s going on when things go right, and what’s going wrong when trials don’t produce desired outcomes. 

Of course, it’s not just hearts that should respond to this type of treatment. There are incredible examples of doctors fully regrowing organs like the esophagus and even the bladder using stem cells. However, so far these have all been one-offs. These treatments are novel and dangerous and are thus only approved by the FDA for patients with few other options.

With that said, the promise of stem cell therapies is hard to ignore. for those who can afford it, domestic regulation isn’t much of an issue. Notable athletes like Peyton Manning have turned to stem cell therapy in moments of dire need. After injuring his neck multiple times and receiving numerous surgeries of no avail, Manning flew to Europe in 2011 to receive an experimental treatment that used stem cells to treat his neck injury. He hasn’t had trouble with his neck since and went on to become one of the most decorated quarterbacks in American football.

Stem cell therapy, if it can realize its potential, will have a massive role to play in the future of medicine. Stem cells are emerging as a fascinating new medical primitive, functioning as a key construction material for the human organism.

The immune system is the body’s natural defense mechanism against disease. However, when our natural defenses aren’t able to eradicate disease-causing agents like cancer cells unaided, cell therapies can be used to turbocharge the performance of immune cells.

In the last decade, immune cell therapies have shown tremendous promise. The most effective immune cell therapy to date has been CAR T-cell therapy, a type of cell therapy capable of treating various blood cancers like B-cell leukemia, lymphoma, and myeloma. CAR T-cell therapy begins with the T-cells, which comprise the body’s defense mechanism against infection, cancer, and other kinds of maladies. T-cells identify and destroy harmful antigens in the body. Frustratingly, for years now, aggressive cancers have succeeded at overwhelming the human immune system and preventing T-cells from removing them. However, in the late 1980s, an Israeli scientist named Dr. Zelig Eshhar wondered if it would be possible to genetically engineer a batch of T-cells to retrofit them with a receptor designed to specifically detect cancer cells. 

Dr. Esharr designed this exact receptor — the CAR or chimeric antigen receptor — in two parts. One part of the receptor called the antigen-binding domain, sticks out from the T-cell’s membrane and sets off an alarm if it detects a targeted antigen. That alarm then activates the second part of the receptor inside the T-cell, called the signaling domain, which instructs the T-cell to attack the antigen it's bound to.

The only problem was that the technology to actually reprogram cells to express these receptors didn’t exist. Though Dr. Esharr helped write the genetic code for this new receptor, it would take decades for our abilities in genetic engineering to catch up with his ideas.

However, in 2010, the first successful demonstration of CAR T-cell therapy finally took place in Dr. Carl June’s lab at the University of Pennsylvania. The patient who underwent the trial therapy, Bill Ludwig, had been suffering from chronic lymphocytic leukemia for a decade before he was enrolled in this historic trial. June’s lab first collected Ludwig’s T-cells and genetically modified them with a CAR specifically designed to look for the markers on cancerous blood cells. Then, these modified T-cells were re-injected back into Ludwig to do their work. Remarkably, just one month after the procedure, Bill Ludwig was cancer free.

The magic of CAR-T therapies, and immune cell therapies in general, is that one dose is often enough to fully cure a patient of a disease. As modified T-cells live on in the patient’s body, they continually hunt for cancerous blood cells and destroy them upon contact. Since 2017, six CAR T therapies have been approved in the United States to treat different forms of leukemia, lymphoma, and myeloma with incredible results. Certain studies show that 90% of patients whose leukemia did not respond to traditional treatments had full remission with CAR T-cell therapy.

However, there are still significant challenges ahead. Though CAR T therapies are very good at fighting blood-based cancers, they’re not yet good at fighting solid tumors — the most common form of cancer by far. Large cancerous masses repeatedly set off T-cells antigen receptors, so much so that after a while, T-cells experience a syndrome called T-cell exhaustion, where their ability to attack the tumor declines. This is one of the reasons it is so difficult for our immune system to naturally fight off cancers. Figuring out how to bolster these therapies to destroy cancerous masses is the next milestone to unlock. Luckily, Carl June is holding out hope that we might reach that milestone in the coming decade.

Other challenges within immune cell therapies involve work on regulatory T-cells. Tregs, as they’re colloquially called, actually help suppress the body’s natural immune response. It’s an important function that ensures that the immune response doesn’t harm the body’s own tissues. All auto-immune diseases, from Type 1 diabetes to multiple sclerosis and even rheumatoid arthritis, are examples of what can happen when this Treg function breaks down. Some of the most exciting recent research in immune cell therapy involves finding ways to genetically enhance the function of Tregs such that destructive auto-immune responses can be tamed. Figuring out how to do this properly could potentially solve the puzzle of curing auto-immune disease once and for all.

Stem cell and immune cell therapies are only scratching the surface of new biotech frontiers scientists are exploring in labs across the world. Another notable approach being pursued intensely is the domain of tissue engineering.

The idea of using cells as therapeutic agents began in the mid-20th century. Arguably, one of the first ‘cell therapies’ has been around since the 1950s, and involves the practice of performing bone marrow transplants in patients with leukemia.

Even though these therapies worked, physicians could only theorize as to why. Outside of macroscopic procedures like bone marrow transplants, the fine structures and inner workings of our bodies eluded us. Luckily, the coming decades were very a productive time in medicine and physiology, with each decade unlocking a critical piece of the cell therapy puzzle. By the end of the 50s, James Watson and Francis Crick discovered the structure of DNA. In the 60s, Ernest McCullough and James Till at the University of Toronto proved the existence of stem cells and produced proof of cells with the ability to differentiate the function. One of the most important decades, however, was the 1970s. That’s when progress in genetic engineering first started to take off. The first genes were isolated in 1973, and by the 1980s, DNA sequencing technology gave us the ability to read and decipher any genetic code. 

Though a few successful techniques of gene editing were developed in the late 80s and 90s, these techniques were limited in their precision and were labor-intensive and time-consuming to produce. The massive breakthrough for cell therapies came in 2012 when Jennifer Doudna and Emmanuelle Charpentier demonstrated that CRISPR-Cas, a mechanism used by bacteria to fight viral infections, could be re-appropriated to edit genes in any cell. The discovery which won Doudna and Charpentier a Nobel Prize showed how a specific RNA molecule could be designed to guide the CRISPR protein to a specific DNA sequence, allowing for targeted gene editing.

Since then, CRISPR has become the go-to tool for genetic editing due to its efficiency and versatility and has unlocked near-limitless potential for the field of cell therapies. If the 20th century was marked by us learning to read and write genetic code, the 21st century, largely on the back of CRISPR technology, will be marked by our ability to actually implement and run the genetic code in real cells. 

Injecting a patient with engineered cells is always risky. As already mentioned, the human immune system is very sensitive, and not always welcoming to the introduction of unfamiliar biological material. To overcome this problem, most cell therapies use the patient’s own modified cells, which is a technique known as autologous cell therapy. Using the patient’s own cells increases the chance that the patient’s organism will accept the cell injection, however, it makes the cell therapy treatment on the whole more time intensive and much more expensive since each patient requires their own unique therapy.
One method of decreasing the cost and increasing the availability of these therapies would be to find a way of producing an off-the-shelf cell therapy that could work for all patients. An off-the-shelf solution would need to solve two big problems. The first problem involves producing enough stem cells or immune cells to be used at scale. The second problem involves figuring out a mechanism such that these mass-produced, modified cells are actually accepted by the patient.
Research is still ongoing on the second question, as companies like Allogene work on strategies to decrease the immune response from allogeneic cell treatment. However, the challenge of scaling cell production might already have a promising solution.

In 2012, scientists John Gurdon and Shinya Yamanaka won a Nobel Prize for their discovery of induced pluripotent stem cells. Essentially, these scientists discovered that through chemical coaxing, common cells like skin cells or blood cells could be transmuted into a pluripotent stem cell state, which means they would have the ability to differentiate into almost any cell type in the body. This technique essentially amounts to a biological time machine that sends differentiated adult cells back into an embryonic state. 

The ability to do this is particularly useful because once skin cells are converted into stem cells, those stem cells can be differentiated into any cell we want, including T-cells, Tregs, or what have you. The discovery of induced pluripotent stem cells, or iPSCs, amounts to the discovery of a mechanism that can convert adult skin or blood cells into any kind of type of cell we need to conduct cell therapies. With this technique, as long as we have abundant blood banks, we should have enough supply to produce the vehicles for cell therapy.

The discovery of iPSCs also solved an intractable ethical dilemma at the center of stem cell therapy research. For a long time, stem cells could only be acquired from the cells present in aborted fetuses. The shaky ethics of harvesting stem cells from these sources resulted in a near-decade-long cooling of stem cell research in the United States, which lasted from 2001 to 2009. During this period, the US federal government abstained from funding any stem cell-related research. Remarkably, with the discovery of iPSCs, even the Vatican has since become a vocal supporter of stem cell research, hosting multiple conferences devoted to promoting the field. 

Another temporary inefficiency in the world of cell therapies is that currently because there are so few FDA-approved cell therapies, therapy designers must also operate as manufacturers for all patients across the country. This makes the manufacture of cell therapies very expensive on a per-unit basis. The situation is equivalent to the early days of the microprocessor industry when chip designers like Intel were also responsible for manufacturing their own chips. Now that the microchip industry has scaled to a sufficient size, it’s much more convenient for firms to specialize in either designing or building chips. As such, most chip design shops now outsource the manufacturing of their chips to specialized, external factories.
In a world where allogeneic or off-the-shelf cell therapy treatments become a reality, a similar shift will need to occur in the cell therapy industry. Luckily, there are already companies working on developing the biomanufacturing technology capable of supplying this need. Mammoth Biosciences and Synthego are just two of the firms working on this problem. Both are leveraging CRISPR technology to build out biological assembly lines where large numbers of cells could be genetically engineered at scale.

Progress in innovative therapeutic treatments can only move as fast as the research progresses. However, there’s currently a big mismatch between the ways clinical trials are conducted and the types of clinical trials we’re trying to run. The FDA, for instance, was established to deal primarily with drugs and devices, and the stages of clinical trials designed to test drug or device-based interventions. Initially, a number of pre-clinical studies are conducted in the lab or on non-human subjects to assess the safety of the treatment. After the pre-clinical steps, there are four FDA-mandated clinical trial phases conducted on human patients, with each trial testing larger and larger patient populations. 

However, when a study concerns the function of modified cells themselves, pre-clinical studies conducted in petri dishes or in mice don’t lend the kind of useful insights that might help these therapies succeed in later-stage clinical trial phases. To date, the most instructive research occurs in very small-scale experimental trials approved for patients in dire conditions. There are, however, proposals that envision a smarter way of experimenting with cell therapies going forward. In the near future, we could use emerging technology like 3D organ printers to supply us with realistic canvases on which scientists could more directly test the efficacy of cell therapies. Already, companies like Organovo are working on 3D printing various human tissues for use in drug testing. 

The success of cell therapies at scale has the potential to transform the field of medicine, leading to a paradigm shift in how healthcare is delivered, reimbursed, and experienced by patients. While there are still challenges to overcome, including regulatory and logistical hurdles, the promise of cell therapies as a revolutionary approach to treating disease is undeniable.

Below are the companies leveraging cell therapy technology to build the future of medicine:

Asimov is building tools to design and manufacture genetically enhanced therapeutics.

Biostage is using a patient’s own stem cells to create organ scaffolds to help with organ replacement.

Vericel is developing cell therapy products for sports-related injuries and severe burns.

Mesoblast is developing allogeneic cell therapy treatments for difficult inflammatory diseases.

Fate Therapeutics is fabricating allogeneic cell therapies to treat a range of immune disorders.

Cellares is developing a turnkey manufacturing solution for cell therapy production.

Deep Genomics is using AI to identify new receptors and genes to design the next generation of cell therapies.

Mammoth Biosciences is building out a biomanufacturing platform leveraging CRISPR technology.

Synthego is designing a CRISPR production platform capable of scaling up the production of cell therapies.

A deep dive into the technologies changing the future of medicine.

Why Cell Therapies Are the Future of Medicine

Each year, roughly 40 billion tonnes of CO2 are emitted into the atmosphere. Given that global energy consumption increases at a rate of 2% every year, it’s likely that our emissions will increase at a similar rate. 

This trajectory conflicts with the Paris Climate Agreement’s 2015 commitment to keep average global temperatures from rising more than 2 degrees Celsius — a goal that climate scientists agree will require net negative emissions. This means that merely reducing existing emissions or switching to less polluting fuel alternatives won’t be enough. We will need a solution to actively take excess carbon dioxide out of the atmosphere.

However, this problem also presents an interesting opportunity. For the past couple of decades, the world has come to look upon atmospheric carbon as a terrible pollutant. However, recent developments in the fields of carbon removal and carbon utilization are increasingly encouraging us to change our lens and instead, look upon carbon emissions as an industrial material in disguise.

If the 20th century was marked by Fritz Haber discovering how to pull nitrogen out of thin air to produce ammonia for use in fertilizer, this might be the century where we learn to pull excess carbon out of the atmosphere to make the commercial materials of the future.

There are two main approaches to removing carbon from the atmosphere. The first is called carbon capture. Although this sounds like an all-encompassing term, it technically only applies to processes that capture carbon emissions directly from power plants. To do this, power plants are retrofitted with systems that capture the exhaust gas generated by the power plant. The carbon in the exhaust gas is then compressed and transported to be processed or stored.

Though carbon capture is an important technology to reduce the amount of carbon dioxide accumulating in the atmosphere, power plants only account for 25% of total annual carbon emissions. The lion’s share of carbon emissions (roughly 60%) are produced by agriculture, transportation, and heavy industry (e.g. metallurgy plants or chemical refineries). The only way to remove carbon emitted from these sources is by directly removing carbon from the air – a strategy known as direct air capture.

Extracting carbon dioxide right out of the air was an idea first suggested by a physicist named Klaus Lackner in 1999. He realized that in the same way a sponge can absorb liquid, there exist other materials that are particularly well-suited to absorb gases — including greenhouse gases. His basic idea was to use fans to suck in air containing carbon dioxide and blow it over materials called sorbents that would attract the CO2.

Over the next few years, Lackner was one of the first to build a direct air capture prototype like this. A few companies like Climeworks, which were inspired by his ideas, soon followed suit. But given the small scale of these projects, the cost of operating them was astronomically high — $600 per ton of CO2 collected. Given that the ideal goal of direct air capture is to remove most of the excess emissions discharged annually, $600 per ton would put the annual cost of doing so at around $14 trillion. 

However, direct air capture started looking a lot more promising when a Canadian company called Carbon Engineering published a paper championing a new direct air capture technique. When operating at scale, this new technique promised to decrease the cost of capture to just $94-232/ton and eventually, capture up to one million tons of CO2 per year.

Though this would be an impressive achievement, it would still only cover the cost of getting the carbon out of the air. To make paying such a price worth it, however, numerous companies have begun investigating methods of carbon utilization— the process of turning carbon collected by direct air capture into useful products like synthetic natural gas or small-chain hydrocarbons. In fact, as part of one of its pilot plants, Carbon Engineering built a mechanism that could turn captured carbon into compounds that could eventually be turned into gasoline, diesel, and even jet fuel. 

Of course, removing carbon to produce more combustible fuels is only carbon-neutral, meaning it doesn’t release extra carbon into the atmosphere. To really mitigate the effects of climate change, a net-negative carbon strategy would be needed; one which would store carbon in geological or deep-ocean deposits for hundreds of years. 

Carbon Engineering already has a scheme like this in the works. It’s collaborating with another company called Storegga Geotechnologies to build Europe’s largest direct-air-capture plant and store the captured carbon beneath the floor of the North Sea. The plant is expected to become operational in 2026, and the company intends to sell the sequestered carbon as carbon credits. However, concerns still abound regarding the environmental impact of placing massive carbon stores at the bottom of our largest bodies of water.


Carbon cycles through the environment naturally. Before there was any life on Earth, carbon rotated from under the crust of the Earth up to the surface and eventually into the atmosphere through volcanic eruptions and tectonic shifts. The carbon would then collect in the oceans, mixing with water to form carbonic acid before settling back down in deep deposits.

The first living organisms on Earth were able to capture carbon and convert it into useful materials much faster than these slower geologic processes. Ever since then, carbon has been a key ingredient in all forms of life, so much so that the carbon cycle can be looked on as a mirror to the cycle of life itself.

The distribution of carbon between the earth’s crust, the oceans, the atmosphere, and the biosphere remained constant for millions of years. This all changed around 1750, the start of the industrial age, when humans realized that burning carbon-based fuels was a bountiful source of energy that could power all kinds of important manual and industrial activities. 

Since then, roughly 5% of the 4000-6000 gigatonnes of all the carbon stored in Earth’s fossil fuel deposits have been burned, releasing roughly 280 gigatonnes of carbon into the atmosphere. Other important changes occurred too, like deforestation required for mass agriculture and industrial operations. This changed the way land was used and reduced the amount of vegetation capable of capturing atmospheric carbon, resulting in a further 150 gigatonnes of net carbon added to the atmosphere.

A deep dive into the technologies turning excess atmospheric carbon into valuable industrial inputs.

Blog

Lidar and the Future of Computer Vision

The Rise of Robots

Investing in Replit

Welcoming Sachin Maini

Why Cell Therapies Are the Future of Medicine

Industrializing Carbon Removal

The Acceleration of Artificial Intelligence

Welcoming David Oko

Welcoming Skye Nguyen and Akshay Chandramouli

The Satellite Renaissance

The Challenge of Building Better Batteries

Investing In Recurrency