Constrained Reinforcement Learning for Unstable Point-Feet Bipedal Locomotion Applied to the Bolt Robot

Constant Roux1, Elliot Chane-Sane1, Ludovic de Matteïs1,
Thomas Flayols1, Jérôme Manhes1, Olivier Stasse1,2, Philippe Souères1
1 LAAS-CNRS, Université de Toulouse, France, 2 Artificial and Natural Intelligence Toulouse Institute, France.
Paper arXiv Video Code Logs (coming soon)
This paper has been accepted for the 2025 IEEE-RAS International Conference on Humanoid Robots (ICHR 2025).

Our framework enables robust and agile locomotion on the bipedal, point-foot, arm-less robot Bolt demonstrating effective control despite its inherently unstable morphology.

Abstract

Bipedal locomotion is a key challenge in robotics, particularly for robots like Bolt, which have a point-foot design. This study explores the control of such underactuated robots using constrained reinforcement learning, addressing their inherent instability, lack of arms, and limited foot actuation. We present a methodology that leverages Constraints-as-Terminations (CaT) and domain randomization techniques to enable sim-to-real transfer. Through a series of qualitative and quantitative experiments, we evaluate our approach in terms of balance maintenance, velocity control, and responses to slip and push disturbances. Additionally, we analyze autonomy through metrics like the cost of transport and ground reaction force. Our method advances robust control strategies for point-foot bipedal robots, offering insights into broader locomotion.

Video

Evaluation across performance, autonomy, and robustness

Additional results

BibTeX

@inproceedings{roux2025bolt,
      title={Constrained Reinforcement Learning for Unstable Point-Feet Bipedal Locomotion Applied to the Bolt Robot},
      author={Constant Roux and Elliot Chane-Sane and Ludovic de Matteïs and Thomas Flayols and Jérôme Manhes and Olivier Stasse and Philippe Souères and Nicolas Mansard},
      booktitle={2025 IEEE-RAS 24rd International Conference on Humanoid Robots (Humanoids)}, 
      year={2025}
}