Cao Wei on Humanoid Robots, the Uncanny Valley and Commercialization

Zhang Xiaojun speaks with investor Cao Wei about Tesla’s humanoid robot, applications and social implications. Originally released in 2022, the episode reflects technology and investment judgments of that period.

Cao reviews humanoid robotics and distinguishes experimental platforms that demonstrate movements from products that reliably complete useful work. A performed movement is a long way from reliable operation in a complex environment.

He treats walking and gait control as foundational problems. A biped must balance, perceive its surroundings and move over changing surfaces before it can reliably perform useful tasks.

Even a robot that can move must still manipulate objects and interact with people. Cao emphasizes that human interaction and changing environments are much more complex than basic locomotion.

Discussing Tesla, he thinks computing, vision and data experience from autonomous driving may help, but robots face body-control and manipulation problems that cars do not.

Beyond vision, a robot may need hearing, touch and force feedback. Perception must identify objects and provide sufficiently accurate information for physical contact.

He uses fine manipulation to illustrate the difficulty: grasping, positioning and applying force demand precision. Moving a box differs from handling soft, irregular or fragile objects.

Discussing Tesla AI Day, he focuses on body proportions, gait and the capabilities shown. His comments describe a demonstration at the time, not proof of a later product’s performance.

Cao rejects the idea that humanoids will necessarily transform every application. He argues that designers should start with the task and then choose the most suitable form.

In factories, a specialized arm or automated forklift may suit a particular workflow better. Such machines are designed around specific problems and do not need human form for its own sake.

A human form may be useful in spaces designed for people, such as when using existing facilities or performing varied tasks. Whether two legs and hands are worthwhile still depends on cost, efficiency and reliability.

They discuss the uncanny valley: a machine that looks almost human but not quite natural may unsettle some people. Cao thinks acceptance of appearance and behavior matters especially in homes.

On when household humanoids might become widespread, Cao mentions widely varying estimates. Those timelines are speculation, not an established product roadmap.

He argues that many users want chores done rather than a machine that resembles a person. Specialized devices such as robot vacuums can already address some needs with less complexity.

Conversation is not the same as doing household work. Examples such as preparing a fish, clearing a blockage or repairing a wall involve varied environments, dexterity and safety judgment.

The speakers use science fiction, including Westworld, to discuss Musk’s long-term ambitions and the relationship between intelligence, bodies and human identity. This is speculative and philosophical discussion, not a technical forecast.

Cao returns to the idea that tasks determine form: where a humanoid body is unnecessary, a simpler structure may work better. He also sees software and cloud capabilities as part of the system.

The conversation turns to service robots. Cao distinguishes machines handling a transport step from systems that directly shape the customer experience; value must be assessed across the whole workflow.

He describes investment experience in cleaning and delivery robots. Transporting items down a corridor may not cover a server’s whole job, so the actual cost saved needs to be measured.

He mentions commercial settings such as delivery in large venues and electric-vehicle charging. Space, task standardization and labor costs all affect feasibility.

Cao says perception and autonomous mapping let some mobile robots move beyond fixed tracks or QR-code routes. Actual capability still depends on the environment, sensors and system design.

He describes soft robotics: flexible materials and structures may help grip fragile or irregular objects, but introduce their own control and engineering challenges.

On investing, Cao says early teams must connect their technology to specific customer needs. He cites cleaning robots as an example where the business and revenue were still early at the time of investment.

They discuss Boston Dynamics’ humanoid robots and dexterous hands. Cao regards dynamic movement as impressive while treating fine hand manipulation as a separate challenge.

Discussing Boston Dynamics’ changes of ownership, Cao considers the gap between advanced research and commercialization. A strong demonstration does not automatically determine product, market or financial outcomes.

He believes service-robot companies may need hardware, software, supply-chain and user-experience capabilities at once, crossing traditional industry boundaries.

Science fiction offers different ways to think about human–robot relationships. They discuss potential negative externalities while separating fictional scenarios from unresolved real-world questions.

Cao sees a long-term opportunity for robotics in China, citing aging, manufacturing demand and clusters of suppliers and customers. The market shares and growth estimates he cites are historical estimates, not current statistics.

He argues that proximity to suppliers and customers can shorten communication, logistics and iteration cycles, and mentions industrial policy. Turning those advantages into success still depends on product quality, costs and demand.

Finally, they discuss automation and jobs, robots learning personal behavior, and the possibility of regulation lagging technology. Cao offers his views, while the social outcomes remain uncertain.

The episode ends by weighing technical opportunity alongside social effects. Cao’s investment outlook and adoption timelines should be understood in their 2022 context.