MoE-Loco: Mixture of Experts for Multitask Locomotion
This work addresses multitask locomotion for legged robots, offering a novel method to handle diverse terrains and gaits, but it appears incremental as it builds on existing MoE techniques.
The paper tackled the problem of multitask locomotion for legged robots by developing MoE-Loco, a Mixture of Experts framework that enables a single policy to handle diverse terrains and gaits, resulting in improved training efficiency and performance through mitigation of gradient conflicts.
We present MoE-Loco, a Mixture of Experts (MoE) framework for multitask locomotion for legged robots. Our method enables a single policy to handle diverse terrains, including bars, pits, stairs, slopes, and baffles, while supporting quadrupedal and bipedal gaits. Using MoE, we mitigate the gradient conflicts that typically arise in multitask reinforcement learning, improving both training efficiency and performance. Our experiments demonstrate that different experts naturally specialize in distinct locomotion behaviors, which can be leveraged for task migration and skill composition. We further validate our approach in both simulation and real-world deployment, showcasing its robustness and adaptability.