Famed robotics expert Rodney Brooks has a blunt message for those investing heavily in humanoid robot startups: your investments are being squandered.
Brooks, a co-founder of iRobot and longtime MIT professor, is especially doubtful of efforts by companies like Tesla and Figure that attempt to teach robots manual skills by having them watch videos of people performing tasks. In a recent essay, he describes this method as “complete wishful thinking.”
The main issue? Human hands are remarkably advanced, containing roughly 17,000 unique touch sensors—something no robot can currently replicate. While machine learning has revolutionized areas like speech and image recognition, those advances relied on years of technology for gathering suitable data. “We lack a similar foundation for tactile data,” Brooks observes.
Safety is another concern. Full-sized bipedal robots require immense energy just to remain balanced. If they topple, they pose a real hazard. Due to physics, a robot twice as large as current models would unleash eight times more destructive force if it fell.
Brooks foresees that in 15 years, the most effective “humanoid” robots will likely feature wheels, several arms, and specialized sensors, moving away from a human-like appearance. For now, he is certain that the billions being spent are going toward costly training projects that won’t ever reach large-scale manufacturing.
Brooks has frequently challenged the optimism of bold founders and enthusiastic investors. Last year, he spoke extensively with TechCrunch about how the hype around generative AI outpaces its actual capabilities, and in some cases, even increases workloads.
For instance, the AI research group METR reported this summer that it enlisted 16 top-rated developers from major open-source platforms to evaluate how AI tools affect real-world software development. The developers were given nearly 250 genuine issues to solve, both with and without AI assistance, and their screens were monitored. When using AI tools, the developers actually took 19% longer to finish their assignments. Interestingly, they believed the AI had made them 20% faster.
Brooks has also maintained that AI does not represent the existential risk that some, including Elon Musk, have warned about. TechCrunch interviewed Brooks on this topic back in 2017 at MIT, when the environment was different but not entirely unlike today’s.
At that point, Brooks noted he was beginning to see more businesses focused on creating datasets for machine learning—a trend that has only grown. He also argued that it wasn’t inevitable for Big Tech to dominate robotics, despite their apparent advantage in data. Yet, leading robotics firms today still find themselves influenced by these tech giants.
Apptronik, a company developing humanoid robots and having raised close to $450 million, includes Google among its investors and joined forces with Google’s DeepMind robotics division late last year to “merge top-tier artificial intelligence with state-of-the-art hardware and embodied intelligence.”
Figure, another prominent AI robotics startup working on humanoid robots, is partially funded by Microsoft and the OpenAI Startup Fund, and in February 2024, it teamed up with OpenAI to blend OpenAI’s research with its own “deep expertise in robotics hardware and software.” The partnership ended almost exactly a year later, in March, with FigureAI announcing it had achieved a “significant breakthrough” in its proprietary, end-to-end AI for robotics and would continue on its own.
