Humanoid Robots are Finally on the Way
Over the past three years, there has been an explosion of companies racing to integrate the hardware and software that will constitute a viable humanoid robot. The new proliferation of innovators is exactly what a futurist would expect to see based on analogies to the launch of automobiles, airplanes, and personal computers. For example, GTP-4 creator OpenAI is investing in this sector.
One of OpenAI’s partners is Halodi Robotics, creator of 1X Technologies. 1X announced the close of a $23.5 million Series A2 funding round on March 23rd, led by the OpenAI Startup Fund. To date, Halodi has made significant progress with its Eve robot, a humanoid on a wheeled base. But a wheeled robot is a limited one, in terms of the human tasks it can take over, and 1X is now focused on a bipedal platform called Neo, to "explore how artificial intelligence can take form in a human-like body."
Neo uses electric actuators rather than hydraulics, runs a screen for a face, and uses humanoid hands. Similarly, Elon Musk announced Tesla's entry into humanoid robotics in 2021, and the company now appears to have made pretty solid progress on the Optimus prototypes. In May, it was shown in a video walking and picking up stuff, and that was pretty much everything it was able to do at the time.
On its most recent earnings call, Musk said the company had made only 5-or-6 Optimus prototypes because the company has “actuator supply issues.” More specifically, Tesla realized that there were no off-the-shelf actuators that "worked well" for Optimus. As a result, Tesla had to design and build the actuators, which create motion by converting electricity into mechanical work. As Musk said, “we've actually had to design our own actuators that integrate the motor, the power electronics, the controller, and the sensors. And really, every one of them is custom designed.
Then, of course, we'll be using the same inference hardware as the car." Then he added, "they're not only lighter, tighter, and more capable than any other actuators that exists in the world, but actually manufacturable in volume." According to Musk, Optimus is “probably the least understood or appreciated part of what we're doing at Tesla," and " it will probably be worth significantly more than the car side of things long-term."
In addition to these established AI players, there’s entrepreneur Brett Adcock, who founded the Vettery "online talent marketplace" and Archer Aviation, one of the leading contenders in the emerging electric VTOL aircraft industry. He recently announced that his latest venture is focused on humanoid robots.
Adcock’s “Musk-like master plan" for his robot company, called Figure, involves three major steps:
As a two-legged humanoid, it'll eventually be able to climb stairs, walk around, and generally operate in most of the environments which humans can. With humanlike hands and body mechanics, it should be able to use the tools we use and do a variety of our jobs.
According to an interview with IEEE Spectrum, the company had more than 40 engineers on board in March 2023, with key talent from the Florida Institute for Human and Machine Cognition, Boston Dynamics, Tesla, Waymo and Google X, among others. So, while the company doesn't seem to have anything built yet, it's a serious venture proceeding at top speed.
Another exciting startup is Apptronik whose latest "general purpose humanoid robot" powered by artificial intelligence is called Apollo. Apollo is a 5-foot-8-inch, 160-pound robot intended to help out people in industrial, retail and manufacturing settings. Apollo is able to lift and carry a payload of 55 pounds and can run for about four hours on its swappable and rechargeable battery. The battery can be changed out so that it can operate for a 22-hour workday.
Apptronik sees robots, like Apollo, not as a replacement for the human workforce but as a tool to be utilized to complete and fulfill tasks. Apptronik looks to tackle labor shortages, reduce turnover, improve efficiency and reduce work-related injuries with robots like Apollo. A Canadian entry called Sanctuary AI, has created a humanoid robot called Phoenix, which is aimed at doing real world tasks in many businesses. Sanctuary has a system for training Phoenix on specific tasks.
In partnership with a business, it films a particular task being done and then digitizes the whole event. That data is used to create a virtual environment, which contains all the objects, and simulates the physics, including gravity and resistance. The AI can then practice the task in that virtual environment. It might perform a million attempts, and when the developers think the AI has mastered the activity in the virtual world, it will be allowed to try in the physical world.
In this way, Phoenix has already been trained to do about 20 different job roles. Sanctuary sees this as an efficient way for humanoid robots to master specific tasks that will be useful for business. By comparison, a robot that can do household chores is much further down the line. Going forward, one of the biggest challenges is to give Phoenix a reliable sense of touch, so it knows how much pressure to apply to an object.
In terms of getting robots into users’ hands, the current leader appears to be Agility Robotics, creator of the groundbreaking bipedal robot Digit. In September, Agility revealed that the company is opening RoboFab™, a robot manufacturing facility in Salem, Oregon with the capability to produce more than 10,000 robots per year. Initial construction of Agility’s 70,000 square foot robot factory began last year, and it is set to open late in 2023.
At full capacity, RoboFab will employ more than 500 human workers in Salem, in addition to employees at Agility’s other locations. Digit will also work in the new factory, in a similar capacity to Agility's customer sites where it’s intended to load, unload, and move totes. Digit is designed from the ground up to go where people go and do useful work, safely, in spaces designed for people. Because so many tasks are designed around human workflows, Digit’s human-centric design enables multi-purpose utility. Initial applications include bulk material handling within warehouses and distribution centers.
Customers that are participating in the Agility Partner Program can expect delivery of the first Digits in 2024, with general market availability in 2025. These humanoid robots are intended to work in human-friendly environments like factories, hospitals, hotels, retail stores, and warehouses. And experience gained here will pave the way for achieving the price-performance levels eventually needed for household applications.
Consequently, as both Musk and Adcock point out, whoever gets the first general-purpose humanoid capable of doing a decently broad set of tasks, will be in an amazing position to generate nearly unlimited value. As such this opportunity space could well become the site of the next high-tech gold rush and the number of entrants is sure to grow.
Given this trend, we offer the following forecast for your consideration. First, personal humanoid robots will not become economically important until they reach the “plateau of productivity” between 2035 and 2040. The rapid proliferation of robot ventures indicates that this technology is rapidly climbing toward the unavoidable “peak of inflated expectations.”
As indicated by the most recent Gartner Hype Cycle for Mobile Robots & Drones (shown in the printable issue) other types of robots including delivery drones, non-humanoid warehouse robots, and even autonomous vehicles are expected to see commercialization by 2033. However, humanoid robots will only realize their potential once many factors related to cost, functionality, and user acceptance are resolved. And this may take up to 20 years.
Second, adoption of humanoid robots will depend primarily on cost, functionality, and reliability. Late in 2022, Goldman-Sachs created four possible scenarios for humanoid robot adoption over the next 13 years largely targeted at factory & warehouse applications in the United States, as well as elder-care applications, globally. Their “base case” involved introducing a functional humanoid robot in 2025 at $250,000 per copy. That estimate was based on the general characteristics of Tesla’s Optimus robot prototype unveiled in September 2022.
To identify initial use cases, Tesla says Optimus will be trained in the factory this year. The average hourly wage for a Tesla factory worker working 8 hours a day is $23.75. By contrast, Optimus is assumed to work 20 hours a day, with 4 hours a day reserved for charging & maintenance. Under this base scenario, Optimus could reach a payback-period of two years for 2025- to-2026. That indicates at least some economic feasibility for commercialization. So, the biggest shortterm question is, “What proportion of tasks within an automobile factory will such a humanoid robot be able to perform as well or better than a human worker?”
Assuming that such robots can perform a wide range of jobs, the other major variable is cost. Given the increasing global labor shortage due to demographics between now and 2035, Goldman-Sachs assumes the labor market demand reflected in its “Global Humanoid Robot Base Case,” its “bull case,” and its “blue-sky case” will depend entirely on the robot’s cost.
Under the “base case,” the unit cost of $250,000 is based on the Optimus bill of materials, and assumes Tesla purchases mid-to-high-end components at market prices for small volumes in 2025; after that a 15% per year cost reduction is assumed through 2035. Under the “bull case,” a 2025 unit-cost of $50,000 was calculated assuming Tesla purchased the components at a much cheaper price by leveraging its broader vehicle procurement team and that it attained the same 20% annual cost reduction documented for Tesla car production.
Given the assumed elasticity of demand, the bull case would enable Tesla to sell one million units a year by 2035. Goldman Sach’s “blue-sky case” assumes a $20,000 per unit cost, which almost no one except Musk believes, so the Trends editors don’t take this scenario very seriously.
Third, even if humanoid robots can achieve extraordinary price and performance targets, non-economic factors are likely to prevent them from making a major dent in the OECD’s labor shortage by 2035. As recently highlighted by AEI economist James Pethakoukis, issues related to trust, safety and privacy are among the hard-to-quantify impediments to rapid adoption.
Fortunately, some settings and cultures are more psychologically amenable to early adoption of humanoid robots. For instance, American factories, which extol “innovation,” are likely to become early adopters in the 2020s. Meanwhile Japan has the best history of accepting such leading-edge innovations in the consumer domain. Once they prove themselves in such environments, humanoid robots will diffuse into the global economy when and if the economics make sense.
Fourth, within the next 10 years, so-called MultiAgent Orchestration (or MAO) platforms will enable multi-vendor robotic solutions to work smoothly together as part of a seamless system. As companies expand their use of robotics, most will eventually have heterogeneous fleets of robots from different vendors performing various tasks, which will require standardized software that can easily integrate to a variety of agents and robot platforms. MAO platforms will reduce the time, effort and cost to onboard new robots and will lower support costs, ultimately making organizations more efficient.
According to Gartner, “this will enable organizations to leverage heterogeneous fleets of mobile robots to assist with more complex activities, delivering cost savings and productivity benefits.”
And, Fifth, well before widespread commercialization, we’ll see a major shakeout among humanoid robot makers, leaving only a few stars to emerge from the “trough of disillusionment.” Just as we saw with automobiles and personal computers, the excitement of a game-changing opportunity will attract a wide array of entrepreneurs and investors. And just as we saw with cellphones, search engines, and social media platforms, we shouldn’t necessarily expect the first leader to be the big winner.