Learning-Based Walking Control and Environmental Perception for Bipedal Robots

Environmental perception plays a crucial role in stable walking control of bipedal robots on complex terrain, especially for low-cost robots lacking external sensors. This paper proposes a reinforcement learning model for bipedal robots that relies solely on self-sensing information for environmental perception. By integrating the robot's self-sensing information and environmental information from the training environment, the model guides the robot to learn reliable environmental perception and stable walking control in complex terrain. The robustness and generalizability of the control strategy were validated through simulation-to-simulation testing.