iRobot (IRBT.US) co-founder Rodney Brooks has emerged as one of the prominent skeptics questioning whether humanoid robots can deliver on the transformative innovation promises made by the industry. Brooks, who is also co-founder of Rethink Robotics (manufacturer of the "Baxter" robot), MIT Panasonic Professor Emeritus, and former director of MIT's Computer Science and Artificial Intelligence Laboratory, believes that the notion of "humanoid robots" working as efficiently as humans within a few years is pure fantasy.
He asserts that "massive investment capital will disappear, and today's humanoid robots will be quickly forgotten by the market." Brooks predicts that truly successful "humanoid" robots will instead adopt wheels, multiple arms, and specialized sensor equipment, with appearances that no longer resemble humans as Tesla CEO Elon Musk firmly believes they should.
In a detailed analysis titled "Why Today's Humanoid Robots Cannot Learn Dexterity," Brooks points out that human hands are extremely complex, possessing nearly 17,000 specialized tactile receptors. Humanoid robots cannot match this capability, as human tactile receptors are sensitive to spatial details down to the precise scale of individual fingerprint ridges.
Brooks identifies walking as another major problem for humanoid robots currently in development. He argues that existing humanoid robots rely on high-gain control and continuously "pumping energy into the system" to maintain upright posture, resulting in high energy consumption and dangerous falls when stability is lost, making them unsuitable for safe deployment in real, crowded environments.
"Current humanoid robots walk nothing like humans. Humans are resilient spring systems that can walk with minimal neural control," Brooks writes. "You can actually see purely mechanical bipedal walking models traverse gentle slopes without any power source, relying solely on passive dynamics and 'stealing' potential energy from downhill walking motions to power the robot mechanically."
Brooks highlights a core risk: since full-sized, walking humanoid robots require enormous energy input to maintain upright posture, they become dangerous when they fall. Looking ahead, Brooks believes successful humanoid robots will actually be equipped with wheels, multiple arms, and specialized robotic sensors, meaning they won't resemble humans in form or appearance.
For investors, this suggests billions of dollars may be invested in humanoid robot forms that Brooks believes are unlikely to reach mass production. "There will be many robots of various forms and appearances for different specialized tasks that humans can perform. But they will still all be called humanoid robots. Meanwhile, substantial funds will have disappeared – funds that may have been spent trying to squeeze performance, any potential performance, from today's humanoid robots. But these human-form-obsessed robots will disappear and be conveniently and quickly forgotten," Brooks warns.
Brooks consistently supports the robotics industry itself but clearly doubts and opposes using "human-like appearance (bipedal + anthropomorphic hands) as the main route," rejecting "human appearance" as a near-term universal solution. As he directly states in his latest article, today's humanoid robots "cannot learn true dexterity," and capital is spending "billions of dollars" on humanoid forms that are difficult to mass-produce. More viable successful robot forms will feature wheels, multiple arms, and specialized sensors, not necessarily resembling humans.
Brooks repeatedly emphasizes that human hands have approximately 17,000 specialized tactile receptors, while current "learning from video" imitation/self-supervised paradigms lack both high-fidelity tactile training data and algorithms and engineering pathways that truly couple tactile feedback with hierarchical subtask planning. Therefore, achieving manipulation and grasping dexterity comparable to humans in the short term is difficult.
Existing full-sized humanoids must continuously "pump energy" into the system (such as relying on ZMP control) to maintain upright posture. Once they lose stability and fall, safety certification for close-proximity work scenarios becomes a hard constraint. Human walking, however, extensively utilizes tendon/muscle elasticity and passive dynamics to recover energy. Brooks considers this a "human-like" nonlinear advantage that current-stage robots cannot replicate.
Brooks also judges that robots capable of large-scale deployment in 15 years will likely have wheels, possibly multiple arms, and use non-human frequency/active light sensing, adopting specialized forms to match tasks and safety costs rather than forcibly replicating human body shape. Current bets on "universal humanoids, plug-and-play human replacements" will create large-scale mismatches between massive inputs and outputs.
Brooks favors specific task + safety compliance-certifiable non-humanoid forms, along with high-end sensing, tactile materials, force control, and planning software that can address dexterity and safety shortcomings. He firmly believes robots will shift from "stunt videos" to reliable MTBF, fall safety, scenario coverage, and unit economics.
Robotics remains a high-growth track globally. As described above, Brooks is not "bearish on robots" but opposes treating "human appearance" as a near-term universal solution. He advocates that funding and market attention should focus more on scalable, practical robot forms and underlying capabilities. His comments serve both as a risk warning about the "humanoid bubble" and a structural bull case for the long-term prospects of the robotics industry.
Jensen Huang has stated multiple times this year that following AI large models, robotics technology, particularly embodied AI, will be the largest potential growth market for the AI chip giant. "Embodied AI" refers to the cutting-edge technology field represented by "AI humanoid robots" that "possess real physical forms and are equipped with edge AI + cloud AI computing systems."
"We have many significant growth opportunities across the company, with AI and robotics being the two largest, representing a total addressable market of trillions of dollars," Huang said when answering shareholder questions at NVIDIA's annual shareholder meeting.
Unlike Tesla, which is fully committed to manufacturing "Optimus" humanoid robots, NVIDIA's recent increased investment in humanoid robots focuses on being a "computing power and platform supplier for embodied AI" – essentially providing "shovels + operating systems" through coordinated cloud training, edge inference, and physical simulation tools. This involves large-scale sales of AI GPU clusters needed for robot training/inference systems + edge computing modules + Isaac simulation and toolchains, serving all humanoid robot players across the industry.
NVIDIA is dedicated to integrating new NVIDIA robotics technology based on 3D omnidirectional dynamic simulation and real-world simulation of the physical world into tech companies' robot model development and training processes. Therefore, NVIDIA's Isaac/Isaac Sim/Omniverse platforms are expected to become industry standards for "simulation + data generation + robot strategy verification," potentially bringing high-stickiness recurring revenue and increasingly strong moats based on integrated hardware and software.
"We are entering an era where the world will soon have billions of robots, hundreds of millions of fully autonomous vehicles, and tens of thousands of super robot factories that can be powered by NVIDIA's software and hardware ecosystem technology," Huang stated.
Comments