RT1复现(一)
环境配置:
conda create -n rt1 python==3.8
1.tensor2robot
1)git clone tensor2robot
git clone https://github.com/google-research/tensor2robot.git
2)modify tensor2robot/requirements.txt
pybullet==2.5.0 --> pybullet
3)pip install
pip install -r tensor2robot/requirements.txt
4)test
python -m tensor2robot.research.pose_env.pose_env_test
5)modify tensor2robot/utils/tensorspec_utils.py
contrib_framework --> tf
- t2r.proto–>t2r_pb2.py
mkdir protoc
cd protoc
wget wget https://github.com/google/protobuf/releases/download/v24.3/protoc-24.3-linux-x86_64.zip
chmod 775 protoc-24.3-linux-x86_64.zip
unzip protoc-24.3-linux-x86_64.zip
pip3 install protobuf==3.20.3
~/protoc/bin/protoc protoc --python_out=`pwd`
2.rt1
1)git clone rt1
git clone https://github.com/google-research/robotics_transformer.git
2)Comment out the last line of robotics_transformer/requirements.txt
注释掉 因为已经编译了
# git+https://github.com/google-research/tensor2robot#tensor2robot
3)pip install
pip install -r robotics_transformer/requirements.txt
4)test action tokenizer
python -m robotics_transformer.tokenizers.action_tokenizer_test
- dataset(后续读取数据集需要用到
pip install rlds tf_agents dm-reverb apache_beam tfp-nightly tensorflow_datasets clu