Humanoid control system improves body posture and position tracking
BeJEPA: A Goal-Conditioned Humanoid Behavior World Model via Unsupervised RL
Robotics
Summary
Controlling humanoid robots so they move naturally and stay oriented in the real world is hard. The paper presents BeJEPA, a new approach that learns to represent a robot’s body and position together, using a type of AI training without labeled goals. It can predict how actions affect the body now and over time, helping the robot recover from falls and reach goals better. The system beats previous methods at keeping track of where the robot is in the world and was tested both in simulation and on a real robot.
What this means in practice
- •For robotics engineers: Improve humanoid robots’ ability to maintain correct body pose and world position while recovering from disturbances or changes in goals.
- •For robot control developers: Build control systems that efficiently guide humanoid robot movement by combining short-term physical predictions with long-term behavioral summaries.
Authors
Liyun Yan, Haoyang Li, Yang Zhang, Jialing Zhu, Zhanxiang Cao, Zihao Zhou, Keqi Zhu, Yizhi Chen, Huang Chuxuan, Shengcheng Fu, Zhong Jiangwei, Qiao Sun, Chen Zhang, Yue Gao
Abstract
Humanoid control requires coordinating whole-body posture with position and heading in the world frame. Mainstream controllers imitate trajectories in the root frame, but struggle to regain the target world pose after displacement. We introduce BeJEPA, a goal-conditioned Behavior World Model learned through unsupervised reinforcement learning from human motion priors and robot interaction. BeJEPA learns a shared representation of body configuration and world placement through a joint-embedding predictive architecture (JEPA). A Dynamic Predictor predicts the next state embedding under an action, while a Successor Predictor summarizes a goal-conditioned policy's expected discounted future state embeddings in a single vector. During joint training, Dynamic grounds the initial representation in observed physical transitions, while Successor increasingly organizes it around long-term behavior. Their composition guides policy improvement through one physical prediction and one behavioral summary, without long imagined rollouts. In 20-second motion-tracking experiments, BeJEPA with online action refinement using Dynamic outperforms all evaluated baselines in global body-position error, reducing it on AMASS by 61.7% relative to BFM-Zero with world-pose inputs and 58.7% relative to SONIC. Physical deployment on a Unitree G1 demonstrates joint attainment of posture, world position, and heading goals, recovery from the ground, and closed-loop return after disturbances and goal changes. Project page: https://ylyem9x.github.io/BeJEPA-project.