Control of Dynamic Parkour Motions for a hopping leg on a Broomstick

As legged locomotion in nature is very dynamic and rich, this elegant art of motion should also be exploited by robots to enable efficient and unrestricted locomotion even on difficult terrain. To attain animal-like dynamic motions for underactuated robots, careful consideration and exploitation of system dynamics is crucial. When dealing with complex parkour environments that require sophisticated traversal, long-term planning becomes essential as movements are interdependent. However, accurate dynamic modeling over extended horizons involves computationally intensive calculations, presenting significant motion planning challenges. To address these challenges, a control system is proposed that optimally plans a feasible path through an obstacle course using motion planning techniques that utilize mixed integer programming (MIP). Subsequently, a state machine executes the contact sequence, employing a PD control scheme and feedforward torques to directly and effectively control the motors in real-time. The proposed approach is tested on the robotic system "Hopping Leg on a Broomstick," demonstrating its autonomous traversal of diverse and challenging parkours. Notably, the controller exhibits robustness even for long control horizons, enabling efficient and reliable locomotion.