Search papers, labs, and topics across Lattice.
This paper introduces MILD, a novel framework for simulating bipedal locomotion on deformable surfaces using a physics-grounded discrete-element contact solver that captures the complexity of foot-terrain interactions. By employing deep reinforcement learning with latent modulation and proprioceptive estimation, the authors train a terrain-aware locomotion controller that significantly outperforms existing methods in generating realistic contact scenarios. Experimental results confirm that MILD enables robots to adapt effectively to varying surface stiffness in real-time, enhancing their operational capabilities in challenging environments.
Robots can now adapt their walking strategies in real-time on deformable surfaces, thanks to a new simulation framework that outperforms existing methods.
Enabling robots to walk on yielding terrain is vital for applications ranging from disaster response to planetary exploration. While bipedal robots hold immense potential, their locomotion on deformable surfaces remains limited as current simulators fail to capture the spatiotemporal heterogeneity of such yielding substrates. We present MILD, featuring a physics-grounded discrete-element contact solver that accurately simulates spatially varying foot-terrain interactions. Complementing this model, we train a terrain-aware locomotion controller via deep reinforcement learning with latent modulation and proprioceptive estimation. Quantitative comparisons against state-of-the-art methods show our approach generates more diverse and realistic contact scenarios during training, resulting in controllers that exhibit natural adaptation on real deformable surfaces. Through hardware experiments, we demonstrate the system's capability for online terrain identification and adaptation across a wide range of surface stiffness.
