Interview with He Xiaopeng (Part-2): Bigger Bets and the Birth of the Humanoid Robot Iron

In this talk

1. 0:11

   Building a different robotics team

   He Xiaopeng says XPeng reconsidered its robotics approach in 2023, retaining fewer than 60 people from a 300-person team before assembling a new group. He wanted people with some experience across AI, cars, engineering, and robotics, rather than relying only on established car or robotics specialists. He says XPeng later hired nearly 80 PhD-level graduates in one department to support long-term technical exploration.

2. 3:29

   Designing a humanoid for human spaces

   He argues that general-purpose robots could serve a broad range of needs, including helping older people, but says they must be safe and comfortable around people. Large quadrupeds can struggle to move in homes, while a small robot may lack useful capabilities; a tall, armored biped can make people wary. XPeng’s current robot is about 1.69 meters tall and may wear clothes or hair, but he says it should not have a face because of uncanny-valley, legal, and social concerns.

3. 15:36

   Taking on motion control and commercialization

   He Xiaopeng identifies whole-body motion control as a key technical challenge, arguing that robots need coordinated control across many joints and actions rather than software that stitches separate movements together. He compares robot motion control to an early stage of automotive development and says XPeng wants AI to coordinate movement more like human instinct. He names hardware reliability, coordination among advanced models, and proving a commercial use case as major obstacles to mass production.

4. 19:02

   Robots as a new company growth curve

   He describes XPeng’s next decade as three growth paths: cars that become capable agents, robots, and global expansion. He believes robots could scale faster than cars once their hardware and software work well, since cars depend on roads, regulations, and gradual production growth. XPeng develops much of its robot hardware in-house, including work on chips and joints, while also relying on partners.

5. 23:31

   Combining capabilities across products

   He Xiaopeng presents the GS as XPeng’s first full-size, six-seat flagship SUV and an example of connecting car, flying-car, and robotics capabilities. He cites redundant systems, steer-by-wire, and integration with the company’s VLA autonomous-driving system as ways to improve control and safety. He also says the car can interpret a request such as fetching water as a sequence of decisions about the user, object, destination, and handoff.