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

arXiv: https://arxiv.org/abs/2606.19633 · HTML: https://mlsanity.com/paper/2606.19633/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in MoE in reinforcement learning and robotics, one of 12 directions in Mixture-of-experts.
- Builds on 25 of its 31 references that are in the map.
- First posted on arXiv 17 Jun 2026.

## Placed in

- Mixture-of-experts / [MoE in reinforcement learning and robotics](https://mlsanity.com/moe/moe-in-reinforcement-learning-and-robotics/)

## Builds on (most cited first, 25 of 25)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning](https://mlsanity.com/paper/2109.11978/) (2021)
- [RMA: Rapid Motor Adaptation for Legged Robots](https://mlsanity.com/paper/2107.04034/) (2021)
- [π*0.6: a VLA That Learns From Experience](https://mlsanity.com/paper/2511.14759/) (2025)
- [Extreme Parkour with Legged Robots](https://mlsanity.com/paper/2309.14341/) (2023)
- [ANYmal Parkour: Learning Agile Navigation for Quadrupedal Robots](https://mlsanity.com/paper/2306.14874/) (2023)
- [Legged Locomotion in Challenging Terrains using Egocentric Vision](https://mlsanity.com/paper/2211.07638/) (2022)
- [Robot Parkour Learning](https://mlsanity.com/paper/2309.05665/) (2023)
- [Learning values across many orders of magnitude](https://mlsanity.com/paper/1602.07714/) (2016)
- [GMT: General Motion Tracking for Humanoid Whole-Body Control](https://mlsanity.com/paper/2506.14770/) (2025)
- [Multi-expert learning of adaptive legged locomotion](https://mlsanity.com/paper/2012.05810/) (2020)
- [Advanced Skills by Learning Locomotion and Local Navigation End-to-End](https://mlsanity.com/paper/2209.12827/) (2022)
- [RSL-RL: A Learning Library for Robotics Research](https://mlsanity.com/paper/2509.10771/) (2025)
- [Learning to Jump from Pixels](https://mlsanity.com/paper/2110.15344/) (2021)
- [Attention-Based Map Encoding for Learning Generalized Legged Locomotion](https://mlsanity.com/paper/2506.09588/) (2025)
- [Track Any Motions under Any Disturbances](https://mlsanity.com/paper/2509.13833/) (2025)
- [CTS: Concurrent Teacher-Student Reinforcement Learning for Legged Locomotion](https://mlsanity.com/paper/2405.10830/) (2024)
- [MoE-Loco: Mixture of Experts for Multitask Locomotion](https://mlsanity.com/paper/2503.08564/) (2025)
- [Learning to Walk from Three Minutes of Real-World Data with Semi-structured Dynamics Models](https://mlsanity.com/paper/2410.09163/) (2024)
- [Mastering Massive Multi-Task Reinforcement Learning via Mixture-of-Expert Decision Transformer](https://mlsanity.com/paper/2505.24378/) (2025)
- [Towards bridging the gap: Systematic sim-to-real transfer for diverse legged robots](https://mlsanity.com/paper/2509.06342/) (2025)
- [Abstracting Robot Manipulation Skills via Mixture-of-Experts Diffusion Policies](https://mlsanity.com/paper/2601.21251/) (2026)
- [VOCALoco: Viability-Optimized Cost-aware Adaptive Locomotion](https://mlsanity.com/paper/2510.23997/) (2025)
- [Toward Reliable Sim-to-Real Predictability for MoE-based Robust Quadrupedal Locomotion](https://mlsanity.com/paper/2602.00678/) (2026)
- [Learning Terrain-Specialized Policies for Adaptive Locomotion in Challenging Environments](https://mlsanity.com/paper/2509.20635/) (2025)

## Abstract (arXiv)

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/ .