Fall-safe bipedal robot enables real-world reinforcement learning

Researchers at the University of Illinois’ Kinetic Intelligent Machine LAB (KIMLAB) have developed a novel bipedal robot platform, called HybridLeg, designed to advance real-world reinforcement learning by enabling safe falls and autonomous recovery. The robot features a unique hybrid leg mechanism combining serial and parallel linkages—a five-bar closed linkage actuated by 12 motors, with most motors concentrated near the pelvis to reduce leg mass and improve dynamic walking performance. This design reduces distal inertia, allowing more accurate physics modeling and efficient, agile locomotion. The robot is fully untethered, integrating onboard computing, sensing, and power systems, and includes a lantern-shaped, sensorized mechanical cover that protects it during falls and facilitates whole-body contact. The HybridLeg platform incorporates a multimodal fall detection system that fuses inertial, proprioceptive, and acoustic sensors alongside an improved stance phase detection algorithm. This enables the robot to detect falls, mitigate impact forces, and autonomously reset to a standing posture after each