Learning to enhance multi-legged robot on rugged landscapes

Navigating rugged landscapes poses significant challenges for legged robots. Multi-legged robots (those with >6 limbs) offer a promising solution for such terrains; recent work [Chong et al, 2023] reveals that such systems can reliably traverse rugose terrain in open loop due to redundancy and mechanical stability. We have also shown that a linear controller, which modulates the vertical body undulation of a 85cm long, 12 legged robot in response to foot-sensed changes in foot contact patterns (related to terrain roughness), improves mobility. However, learning-based approaches have significant potential to improve performance by identifying intelligent gait adaptation strategies to navigate unstructured terrain. We posit that an experimentally validated physics-based numerical simulator for this robot can rapidly advance capabilities by allowing wide parameter space exploration. Here we develop a MuJoCo numerical simulation of the robophysical model and use the simulation to develop a reinforcement learning-based control framework that dynamically adjusts horizontal and vertical body undulation, and limb stepping in real-time. Our approach improves robot performance in simulation, laboratory experiments, and outdoor tests. Notably, our real-world experiments reveal that the learning-based controller achieves a 30% to 50% increase in speed (3 cm/s to 5 cm/s) compared to a linear controller, which only modulates vertical body waves.