Skip to content

论文arXiv 2611.262932 天前

LocoMuJoCo开源全身控制基准:统一CPU/GPU后端与Sim-to-Real部署 ​

本文由 AI 根据公开内容整理,可能有误,请以原文为准:LocoMuJoCo: A Comprehensive Imitation Learning Benchmark for Whole-Body Control(arXiv 2611.26293)

核心内容 ​

LocoMuJoCo是专为模仿学习与强化学习设计的全身控制开源基准库,旨在解决异构计算与数据碎片化痛点。该库通过单一API无缝耦合MuJoCo CPU仿真器与MJX、MjWarp GPU加速后端,覆盖9款人形、3款骨骼肌模型及4款四足机器人形态。项目为每个人形本体提供精选原生动作捕捉数据集,并兼容AMASS与LAFAN1数据集的重定向,代码基于MIT协议在GitHub开源,配套提供Sim-to-Real部署仓库。

技术架构上,LocoMuJoCo采用模块化插件设计,实现观测空间、奖励函数、控制模式、地形生成及域随机化的完全解耦与互换。算法层提供紧凑的JAX实现,支持RL与IL算法的JIT编译端到端训练,显著提升迭代效率。针对跨形态迁移难题,构建了标准化的SMPL-to-Robot及Robot-to-Robot重定向管线。此外,配套的loco-mujoco-real仓库打通了从仿真策略到真机部署的完整链路,为具身算法验证提供标准化基础设施。

关键看点 ​

  • 异构计算统一:单一API封装MuJoCo CPU与MJX/MjWarp GPU后端,消除仿真与加速训练间的接口割裂
  • 多模态数据集成:内置16种机体原生动捕数据,支持AMASS/LAFAN1重定向及SMPL跨形态迁移管线
  • JAX端到端训练:提供RL/IL算法紧凑JAX实现,支持JIT编译加速,配合模块化插件实现组件热插拔
  • Sim-to-Real闭环:配套loco-mujoco-real开源仓库,提供从仿真策略训练到真机部署的标准化工具链

工程与产业观察 ​

该基准通过统一CPU/GPU后端与标准化重定向管线,大幅降低了全身控制算法在不同机体间的迁移门槛。其JAX原生支持与Sim-to-Real配套工具,使其成为验证具身大模型运动能力的关键基础设施,有助于推动人形与四足机器人运动控制研究从定制化开发走向标准化评测。

英文摘要 (Abstract) ​

LocoMuJoCo is an open-source benchmark for imitation learning (IL) and reinforcement learning (RL) on whole-body control. The library couples MuJoCo (CPU) with MJX and the MjWarp backend (GPU) under a single API, exposes 9 humanoid, 3 (musculo-)skeletal, and 4 quadruped embodiments, and ships a curated native motion-capture dataset for every humanoid, to which it can also retarget AMASS and LAFAN1. A modular plug-in architecture makes observations, rewards, control modes, terrains, and domain randomization fully interchangeable, while compact JAX implementations of many RL and IL algorithms provide JIT-compiled end-to-end training, and the companion loco-mujoco-real repository provides sim-to-real deployment. We describe the design, the SMPL-to-robot and robot-to-robot retargeting pipeline, and a usage example. LocoMuJoCo is released under MIT at https://github.com/robfiras/loco-mujoco.

作者:Firas Al-Hafez, Davide Tateo, Oleg Kaidanov, Aditya Bhatt, Artur Zelik, Zhengyi Luo, Jan Peters

基于本页技术方向关联的整机参数、核心零部件、实测报告与同赛道选型横评。

外部来源

本页事实对应的公开页面,以原文为准。(点击将离开本站)

  1. arXiv 2611.26293:LocoMuJoCo: A Comprehensive Imitation Learning Benchmark for Whole-Body Control ↗访问 2026-11-20