LimX Dynamics’ Biped Robot P1 Conquers the Wild Based on Reinforcement Learning
Based on Reinforcement Learning (RL), LimX Dynamics’ Biped Robot P1 ventured into Tanglang Mountain* in Shenzhen for the first time. With Zero-shot Learning, non-protected and fully open testing conditions, P1 successfully navigated the completely strange wilderness of the forest, demonstrating exceptional control and stability post reinforcement learning by dynamically locomoting over various complex terrains.
*The main peak of Tanglang Mountain is 430 meters in elevation, covering an area of 1,000 square kilometers.
In this test, the greatest difference seen between the forest environment and the laboratory or city is that no two steps, slopes, or even flat grounds are identical; from the bottom of the mountain to the peak, the terrain varies significantly. Soil erosion exposes underlying rocks, slopes are covered with entwining vines, and weathered soil layers become sandy, covering the ground. The ditches built from materials on-site are irregularly shaped. The conditions in the wild pose scenarios P1 has never encountered before. They are also equally challenging for humans to navigate.
During the simulation training, P1 was not fed with any data related to the forests or hiking, and the differences from the deployment environment were vast. Yet, it still managed to adapt to the new surroundings, walking through the unpredictable forest. This success is attributable to LimX Dynamics’ systematic training of reinforcement learning, ensuring that the trained policies are feasible, usable, and reliable in real-world applications.
P1 is LimX Dynamics’ innovative point-foot biped robot, serving as an important platform for the systematic development and modular testing of reinforcement learning. It is utilized to advance the research and iteration of basic biped locomotion abilities. The success of P1 in conquering forest terrain is a testament to LimX Dynamics’ systematic R&D in reinforcement learning.
Beyond locomotion, LimX Dynamics continues to make breakthroughs in manipulation and loco-manipulation on humanoid robots, with more developments to be shared in the future.
