Real-time Replanning and Control of Quadruped Robots for Blind Locomotion on Uneven Terrain

Adaptability to different terrains is crucial for legged robots to perform diverse tasks. This paper proposes a real-time updated control framework that improves the robustness of quadruped robots walking blindly on rough terrain. The proposed framework that considers both stability and kinematics has several advantages: 1) no visual feedback, 2) self-adaptation to unpredictable touchdown position and 3) self-adjustment for unknown rough terrain. The controller tracks the desired robot's CoM (center of mass) postion and posture through optimized distributed ground reaction forces obtained by QP (quadratic programming). By implementation of the proposed method, SIT-Dog robot can achieve stairs climbing and irregular terrrain crossing in the simulation environment.