Sangbae Kim - Physical intelligence and Cognitive Biases Toward AI

In this talk

1. 1:09

   Taking robots beyond the factory

   Factory robots usually work in fixed positions and can rely on precise placement. Outside factories, robots must move, balance, handle unknown surroundings, and interact with objects. Kim describes proprioceptive actuation in Cheetah, Mini Cheetah, and humanoid robots as a way to support that interaction.

2. 4:31

   Why people misjudge AI

   Kim says people often assess AI by human standards and assume systems understand or perceive as people do. We also overlook physical skills we perform without conscious attention: he asks listeners to notice their fingers while reaching into a pocket. He connects this blind spot to the difficulty of general robotic grasping and picking.

3. 6:37

   Abstract language and visible difficulty

   People readily turn phrases like “clean your room” or “wash the dishes” into grounded actions, but Kim argues that this skill is easy to underestimate. He also says people notice a robot backflip more than the balance control behind it: two undergraduates produced the flip in about three days, while balanced model predictive control work took over two years. This reflects Moravec’s paradox: computers have advanced in tasks people consider intellectual, while perception and mobility remain hard.

4. 10:28

   Physical learning needs more than demonstrations

   Kim questions whether large volumes of human demonstrations can teach a humanoid to walk robustly or recover when shoved. He emphasizes that physical systems face execution uncertainty at every moment, and failure has real-world costs. His lab is exploring fast reflex algorithms for manipulation and ways to combine them with data-driven models and reasoning methods such as model predictive control.

5. 15:44

   Making physical intelligence visible

   Asked how to bring more attention to physical intelligence, Kim says it takes deliberate effort because physical capability is a cognitive blind spot. He recommends raising the topic, asking questions, and bringing more hardware into view. He also suggests that reliance on phones and digital technology can reduce attention to physical interaction.