实践Mujoco Playground。包括安装以及对dm_control_suite、locomotion的测试。
GPU:至少24G显存,因为训练时需要18G左右显存。如RTX4090、RTX3090等
操作系统:Linux,如Ubuntu22.04
基础软件:NVIDIA显卡驱动、CUDA12.4、Anaconda(Python虚拟环境)
# 更新系统 sudo apt update # 安装ffmpeg sudo apt install -y ffmpeg# 建立虚拟环境 conda create -n mujoco python==3.12 -y # 激活虚拟环境 conda activate mujoco# clone源码 git clone https://github.com/git-cloner/mujoco_playground_sample # 切换目录 cd mujoco_playground_sample # 安装依赖库 pip install -r requirements.txt \ -i https://pypi.tuna.tsinghua.edu.cn/simple# 解决“untimeError: Unable to load cuSPARSE. Is it installed?”问题 unset LD_LIBRARY_PATH # 测试dm_control_suite CUDA_VISIBLE_DEVICES=0 python dm_control_suite.py # 测试运动 CUDA_VISIBLE_DEVICES=0 python locomotion.py# 官网例子 git clone https://github.com/google-deepmind/mujoco_playground cd mujoco_playground/learning/notebooks # jupyter 转 python jupyter nbconvert --to python dm_control_suite.ipynb # 其他参考文件 https://research.mels.ai/ide?mels=UnitreeGo1.qkazy


