Publications

You can also find my articles on my Google Scholar profile.

Conference Papers

Latent Conditioned Loco-Manipulation Using Motion Priors

Published in IEEE-ICHR, 2025

We present a framework for loco-manipulation in humanoid and quadruped robots that combines imitation learning with latent-space control. Unlike standard Deep Reinforcement Learning approaches focused on single skills, our method trains a multipurpose motion policy capable of acquiring diverse low-level skills while enabling high-level goal control. Extending prior work in character animation, we incorporate safety constraints and a diffusion discriminator to improve motion fidelity. We demonstrate the effectiveness of our approach through simulation on the H1 humanoid and Solo12 quadruped, and validate transferability via deployment on Solo12 hardware.

Recommended citation: Maciej Stępień, Rafael Kourdis, Constant Roux, Olivier Stasse. Latent Conditioned Loco-Manipulation Using Motion Priors, IEEE-ICHR 2025.
arxiv | Project Page

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

Published in IEEE-ICHR, 2025

In this paper, 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.

Recommended citation: Constant Roux, Elliot Chane-Sane, Ludovic De Matteïs, Thomas Flayols, Jérôme Manhes, Olivier Stasse, Philippe Souères. Constrained Reinforcement Learning for Unstable Point-Feet Bipedal Locomotion Applied to the Bolt Robot, IEEE-ICHR 2025.
arxiv | Project Page

Reinforcement Learning from Wild Animal Videos

Published in arxiv, 2024

We present RLWAV, a method to teach legged robots locomotion skills by learning from thousands of wild animal videos. Using a video classifier and reinforcement learning, our approach enables a robot to acquire diverse behaviors like walking and jumping in simulation and transfers these skills directly to a real quadruped, despite the domain and embodiment gap.

Recommended citation: Elliot Chane-Sane, Constant Roux, Olivier Stasse, Nicolas Mansard. Reinforcement Learning from Wild Animal Videos.
arxiv | Project Page

Whole-body MPC and sensitivity analysis of a real time foot step sequencer for a biped robot Bolt

Published in IEEE-ICHR, 2024

This paper introduces a novel whole-body model predictive controller combined with a footstep sequencer for the bipedal robot Bolt, enabling robust locomotion. Simulations show effective velocity tracking and recovery from pushes and slips. Additionally, a theoretical sensitivity analysis of the footstep sequencing problem is provided to deepen result insights.

Recommended citation: Constant Roux, Côme Perrot, Olivier Stasse. Whole-body MPC and sensitivity analysis of a real time foot step sequencer for a biped robot Bolt, IEEE-ICHR 2024.
arxiv