RoboBoom
IEEE Robotics and Automation Society· February 18, 2026· 21:12

IROS 2025 Keynotes - Humanoid Robot Systems: Xingxing Wang

Watch on YouTube

Xingxing Wang of Unitree Robotics discusses humanoid robot design, motion learning, and the challenges of making embodied AI work across tasks and environments at the IROS 2025 keynote forum.

In this talk

  1. 5:27

    Building humanoids for practical use

    Wang says robots need to handle a broad range of tasks and changing conditions to make work easier for people. He describes making robots accessible and useful in everyday life as a company goal.

  2. 9:58

    Designing robots and collecting motion data

    Wang introduces a fourth-generation humanoid and explains that its face was designed in response to requests for a more humanlike appearance. He says reaching human-level task performance requires accounting for errors, and highlights full-body calibration through remote motion capture as important for data collection.

  3. 13:33

    Improving learned movement

    Wang describes progress in reinforcement learning, from basic actions to walking and running, with boxing as a current focus. Boxing presents a different challenge from dancing: the robot must withstand impacts and perform varied combinations of movements.

  4. 16:42

    Testing generative models and embodied AI

    Wang says a video generation model has been used to guide robot movement, but its ability to generalize remains limited. The project’s code and dataset were released publicly, while aligning generated video with robot motion and meeting the computing demands remain challenges. He also points to a lack of evidence that embodied AI generalizes across environments, tasks, and robot designs.

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

More talks