RoboBoom
PyTorch· June 11, 2025· 16:03

Real-World Robotics as the Next Frontier for AI? - Pierre Rouanet, Pollen Robotics

Pierre Rouanet, co-founder of Pollen Robotics, discusses how AI is changing robotics and what it will take to bring useful robots into homes and workplaces, speaking at a PyTorch event.

In this talk

  1. 0:03

    Robots are moving into less controlled settings

    Rouanet draws on about a decade building humanoid robots and other robotic systems. He describes a shift from factory automation, where tasks and surroundings are fixed, toward robots that must adapt and interact in everyday environments.

  2. 2:38

    AI advances perception, interaction, and movement

    Vision and language models help robots recognize varied objects and respond to context; a robot could interpret an instruction to pick up an orange and put it in a brown bag. Deep reinforcement learning paired with realistic simulation has also helped humanoids walk, run, and jump, while narrowing the gap between simulation and real hardware.

  3. 5:50

    Household tasks demand broader skills and longer plans

    Rouanet argues that cleaning a kitchen can be harder for robots than acrobatic movement because objects and room layouts vary, and there is no single correct sequence of actions. A kitchen task may require planning tens of seconds or a minute ahead, beyond the short horizons used for locomotion.

  4. 7:56

    Robotics needs interactive physical data

    He says progress likely depends on more first-person, responsive data that captures physical interaction, rather than relying only on static videos scraped from the internet. Robot fleets and teleoperation can collect this data, but scaling to potentially millions of examples is difficult; simulation and human demonstrations also face transfer gaps.

  5. 11:13

    Open, affordable projects broaden participation

    Rouanet says hardware costs, maintenance, and safety can keep robotics research confined to large labs and companies. He highlights a low-cost manipulator with printable plans, parts lists, compatible software, and a T-shirt-folding tutorial, plus an accessible locomotion robot whose simulated policy can run on real hardware.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks