ANYmal

I used ANYmal as a testbed to investigate energy-efficient legged locomotion: by equipping its knee joints with parallel elastic actuators, I explored how passive elasticity reduces torque requirements and impact loads in walking and stair climbing. arXiv

Moreover, I developed a sim-to-real framework that integrates physics-grounded actuator modeling for ANYmal, using reinforcement learning to produce locomotion policies whose energy cost of transport is ≈ 32% lower than prior baselines — showing robust, efficient walking on the real robot.