Could the world’s “biggest product” turn out to be its biggest fantasy? Rodney Brooks, the co-founder of iRobot and a pioneer in autonomous systems, thinks Elon Musk’s vision of humanoid robot assistants is far from achievable in the near term. “In my opinion, believing that this will happen any time within decades is pure fantasy thinking,” Brooks wrote, dismantling claims that Tesla’s Optimus could equal human capabilities in just a few years.
At Brooks’ core skepticism is a fundamental technical gap: human dexterity and tactile sensitivity remain unmatched by machines. A human hand contains 17,000 mechanoreceptors that give rich, nuanced feedback to the nervous system, allowing it to perform rapid adjustments in grip, force, and motion. Most humanoid robots, including Optimus, can manage only limited tactile sensing, often relying exclusively on inputs from their vision systems. This is also a reason why the lack of fine-grained touch data hampers adaptability in dynamic environments, where real-world manipulation demands continuous micro-adjustments. Even advanced arrays struggle to provide decent durability, contamination, and the fusion of touch data with vision and proprioception-resulting in high grasp failure rates.
Recent demonstrations, such as F-TAC Hand’s full-hand tactile coverage, demonstrate what can be achieved when dense multimodal sensing is combined with adaptive motor control. These systems exist only in a lab setting, and their translation to mass production is far from reality. Integration of this kind of capability into commercial humanoids would require a high degree of engineering precision, along with robust learning frameworks that generalize to a wide range of tasks with high variability and uncertainty-something current imitation-learning pipelines fall far short of achieving.
Brooks also points out some flawed training strategies in companies like Tesla and Figure. Both have moved towards vision-only imitation learning from human video data, assuming that to be the key to unlocking visual precision for dexterity. Lacking any force feedback or high-resolution tactile sensing, these methods risk producing brittle skills that collapse outside the staged environments. Optimus, in Tesla’s own demonstrations, performs basic chores-opening cabinets and tearing paper towels-but only within a controlled setting with known objects and lighting. Continuous operation in cluttered, unpredictable human spaces remains unproven.
Beyond the mechanical and algorithmic challenges, Brooks warns about the environmental impact of AI-powered robots. Training large-scale models for humanoid control requires high computational resources, often deployed in hyperscale data centers using billions of gallons of freshwater and an immense amount of electricity. In 2022 alone, 5 billion gallons of freshwater were consumed by the cooling of Google’s data centers, while that of Microsoft rose 34%. If these are fueled by fossil fuels, they then become huge contributors to greenhouse gas emissions. According to estimates from the International Energy Agency, by 2026, data centers will use as much as 1,000 terawatt-hours per year-equivalent to all of Japan’s electricity.
This would also include efforts toward efficiency in AI models, the use of domain-specific systems rather than huge general-purpose networks, and powering data centers with renewable energy. Innovations in hardware, such as neuromorphic chips and optical processors, have the potential for great energy savings but have been adopted very slowly. As Mahmut Kandemir says, “To make AI sustainable, we need proactive solutions streamlining models, developing greener infrastructure, and fostering collaboration across disciplines.”
Meanwhile, investment in humanoid robotics is warming up to arguably dangerous levels. China’s National Development and Reform Commission has cautioned against a bubble, citing over 150 humanoid robot makers within the country’s borders and a flood of highly similar models. Venture capital reports agree that humanoid robotics is pulling in record deal counts, mostly because of AI hype rather than any kind of clear commercial value. For Daiva Rakauskaitė from Aneli Capital, industrial robots bring in revenues, but humanoids can’t yet prove their commercial value.
Case studies of specialized non-humanoid robots drive Brooks’ point home: wheeled warehouse bots, robotic arms with parallel jaw grippers, and suction-based end effectors deliver real, measurable ROI today because they are engineered for specific, predictable tasks. They avoid the complexity and fragility of trying to replicate the full spectrum of human movement and touch. Such deep technical and operational challenges need to be solved before Musk’s claim that Optimus could create $30 trillion in revenue is realized.
Until humanoid robots match human dexterity, can operate reliably in unstructured environments, and do so with sustainable energy use, Brooks’ forecast of robots long gone and mostly conveniently forgotten is a credible counterweight to the hype.
