有意报考/参与OpenMOSS实验室具身智能/人形机器人智能团队的同学,均需先完成下列入门练习(任务一至三必做,任务四至六选做)。 我们共享出来这个入门练习供广大具身智能/人形机器人智能爱好者作入门参考。
- 会用Google, ChatGPT/DeepSeek和Claude Code/Cursor
- 会用Linux
- 会用Git和GitHub
- 学习传统机器人学中的基础知识,比如基础坐标变换、正逆运动学、动力学、控制理论等;
- 在PyBullet/Mujoco仿真实现基于传统机器人运动控制的机械臂物体抓取;
- 可选:在机械臂真机实现基于传统运动控制的机械臂物体抓取,同时学习机器人操作系统ROS;
备注:
- 参考书目
- Introduction to Robotics: Mechanics and Control, Stanford
- Robotic Manipulation, MIT
- PyBullet仿真链接:https://github.com/bulletphysics/bullet3
- Mujoco仿真链接:https://mujoco.org/
- 闯关游戏:https://rcfs.ch/
- 学习强化学习基础知识,并在OpenAI Gym环境中选择几个感兴趣的tasks使用强化学习方法训练并测试成功率;
- 在PyBullet/Mujoco中训练机械臂的物体抓取强化学习策略;
- 可选:尝试在真机部署强化学习策略,体会Sim2Real的过程;
备注:
- 参考书目
- 《神经网络与深度学习》相关章节
- Introduction to Reinforcement Learning, 2nd & David Silver's UCL Course
- UCB 285 Deep Reinforcement Learning
- OpenAI Gym链接:https://gymnasium.farama.org/index.html
- 复现模仿学习经典baseline方法Diffusion Policy
- 同时学习Huggingface机器人学习框架LeRobot
- 学习并利用现有的VLA大模型如Pi/GR00T等, 基于StarVLA框架探索利用现有的仿真数据集训练VLA机械臂模型;
- 可选:采集真机数据集微调VLA模型并真机部署;
- 了解:WAM相关内容,https://openmoss.github.io/Awesome-WAM
备注:
- Large Models: https://stanford-cs336.github.io
- Pi: https://github.com/Physical-Intelligence/openpi
- GR00T: https://github.com/NVIDIA/Isaac-GR00T
- StarVLA: https://github.com/starVLA/starVLA
- 桌面级任务规划
- 参考ICRA 2023 Outstanding Paper “Code as Policies: Language Model Programs for Embodied Control” 在PyBullet仿真实现基于视觉模块和大语言模型的机械臂物体抓取;
- 可选:在机械臂真机实现基于LLM/VLM的任务规划;
- 了解:Agentic VLA相关内容,CaP-X: A Framework for Benchmarking and Improving Coding Agents for Robot Manipulation;
- 场景级任务规划
- 配置任意一个仿真环境, EAI或者EmbodiedBench,选择一个模型跑通baseline;
- 备注
复现OmniH2O: Universal and Dexterous Human-to-Humanoid Whole-Body Teleoperation and Learning (https://omni.human2humanoid.com/) 的人形机器人运动控制方法, 学习人形机器人基础知识, 仿真训练+Sim2Real的过程
- 全身运动控制
- 全身遥操作
- 全身模仿学习
参考:
- Unitree Robotics GitHub: https://github.com/unitreerobotics
- Hover: https://github.com/NVlabs/HOVER
- Underactuated Robotics from MIT
- 如何做研究
- 会议期刊
- 机器人:Science Robotics, RSS, CoRL, ICRA, IROS, RLC
- 机器学习:ICLR, NeurIPS, ICML
- 计算机视觉:CVPR, ICCV, ECCV
- 自然语言处理:ACL, EMNLP, COLM
- 在线研讨班
- CMU Robotics Institute Seminar: https://www.youtube.com/@cmurobotics
- MIT Robotics Seminar: https://www.youtube.com/@MITRoboticsSeminar
- MIT Embodied Intelligence Seminar: https://www.youtube.com/@mitembodiedintelligence8675
- Stanford Seminar: https://www.youtube.com/@stanfordonline
感谢上述相关work开源其代码!






