Step to Step Dynamic

Step-to-step (S2S) dynamics models analyze the discrete transitions between consecutive steps in locomotion, aiming to improve the robustness and efficiency of robot control and reinforcement learning algorithms. Current research focuses on developing more accurate multi-step prediction models, mitigating compounding errors through techniques like any-step dynamics models and multi-timestep objectives, and incorporating these models into control strategies using methods such as barrier functions and system-level synthesis. This work is significant for advancing the capabilities of legged robots in challenging environments and improving the sample efficiency of reinforcement learning, with applications ranging from bipedal walking to more complex multi-contact locomotion.