2 April 2025 in IEEE Robotics and Automation Letters
The complex nonlinear nature of friction in real-world applications renders traditional physical models inadequate for accurately capturing its characteristics. While numerous learning-based approaches have addressed this challenge, they often lack interpretability and fail to uphold the physical guarantees essential for reliable modeling. Additionally, existing structured data-driven methods, despite their efficacy in handling nonlinear systems, seldom account for the specific traits of friction or ensure passivity. To overcome these limitations, we introduce a structured Gaussian Process (GP) model that adheres to the physical consistency of joint friction torque, enabling data-driven modeling in function space that accurately captures Coulomb and viscous friction characteristics while further guaranteeing passivity. We experimentally validate our approach by deploying the friction model on a two-degree-of-freedom (2-DoF) leg prototype. Our approach exhibits robust performance in the presence of non-passive and high-noise data. Experimental results demonstrate that our joint friction model achieves enhanced data efficiency, superior friction compensation performance, and improved trajectory tracking dynamics compared to other friction models.
19 March 2026 in IEEE Robotics and Automation Letters
We propose a contact-explicit hierarchical architecture coupling Reinforcement Learning (RL) and Model Predictive Control (MPC), where a high-level RL agent provides gait and navigation commands to a low-level locomotion MPC. This offloads the combinatorial burden of contact timing from the MPC by learning acyclic gaits through trial and error in simulation. We show that only a minimal set of rewards and limited tuning are required to obtain effective policies. We validate the architecture in simulation across robotic platforms spanning $\mathbf {50}\,$kg to $\mathbf {120}\,$kg and different MPC implementations, observing the emergence of acyclic gaits and timing adaptations in flat-terrain legged and hybrid locomotion, and further demonstrating extensibility to non-flat terrains. Across all platforms, we achieve zero-shot sim-to-sim transfer without domain randomization, and we further demonstrate zero-shot sim-to-real transfer without domain randomization on Centauro, our $\mathbf {120}\,$kg wheeled-legged humanoid robot.