CTS-MoE: Implicit Terrain Adaptation via Mixture-of-Experts for Perceptive Locomotion

2026-06-17

AI Summary

This work addresses the challenge of adaptive locomotion for quadrupedal robots on discontinuous terrains, such as stairs and gaps, by proposing the CTS-MoE method, which balances shared foundational capabilities with mitigation of multi-task reward interference. CTS-MoE uniquely integrates a dense mixture-of-experts architecture with perception-gated routing, enabling implicit terrain adaptation through a perception-driven mechanism—without requiring high-level selectors or explicit terrain classifiers. To prevent value interference, the approach employs a multi-critic structure with task-specific value heads and leverages a single-stage concurrent teacher-student framework for end-to-end training. Evaluated on the Unitree Go1 platform in both simulation and real-world experiments, the method significantly reduces trajectory tracking error and improves traversal success rates on both seen and unseen terrains compared to monolithic policies.

Abstract

Perceptive legged locomotion over discontinuous terrain (e.g., stairs, gaps, and obstacles) requires adaptive behavior, as a single conservative gait cannot produce the anticipatory maneuvers needed for abrupt topology changes. Cast as multi-task reinforcement learning, this problem introduces a tension between sharing and separation. Tasks use a common locomotion base but have conflicting rewards, so a policy must share behavior while avoiding value interference. Prior work addresses only one side, with monolithic policies sacrificing specialization and hierarchical sub-policies sacrificing generalization across transitions and unseen terrain. We propose CTS-MoE, which combines a dense mixture-of-experts actor with perception-based gating to compose shared behaviors and a multi-critic with task-specific value heads to prevent interference. The model is trained end-to-end in a single-stage concurrent teacher-student setup that handles partial observability and avoids sequential distillation, with task labels used only during training. At deployment, routing depends solely on perception, allowing terrain adaptation without a high-level selector or terrain classifier. Experiments on a Unitree Go1 in simulation and on hardware across seen and unseen terrains show task-aware specialization, with lower tracking error and higher success rates than monolithic baselines. Project Website: https://cts-moe.github.io/ .

Problem

• perceptive locomotion
• discontinuous terrain
• multi-task reinforcement learning
• value interference
• terrain adaptation

Innovation

• Mixture-of-Experts
• Perceptive Locomotion
• Multi-task Reinforcement Learning
• Implicit Terrain Adaptation
• End-to-end Training