CTS-MoE: Implicit Terrain Adaptation via Mixture-of-Experts for Perceptive Locomotion
Researchers have developed CTS-MoE, a novel approach for perceptive legged locomotion that utilizes a mixture-of-experts model combined with perception-based gating. This system enables robots to adapt their gait and behavior in real-time to discontinuous terrain, such as stairs and gaps, without requiring a separate high-level selector or terrain classifier. Experiments on a Unitree Go1 robot demonstrated that CTS-MoE achieves lower tracking error and higher success rates compared to monolithic baseline policies, showcasing its effectiveness in both simulated and real-world environments. AI
IMPACT This research could lead to more robust and adaptable robotic systems capable of navigating complex, real-world environments.