The Official Code for "Object-Oriented Option Framework for Robotics Manipulation in Clutter", which is published at IROS 2023.

O3F

|- gym: modified gym based on our needs

|- models: directory for saving by-product models

|- stable_baselines3: modified stable-baselines3 (sb3) based on our needs

|- train_oe.py: code for Option Executor (OE) training

|- train_op.py: code for Option Planner (OP) training

1. Please finish the following steps to install conda environment and related python packages
   • Conda Environment create

   conda create --name <env_name> --file spec-list.txt

   • Package install

   pip install -r requirements.txt

2. The environments used in this work require MuJoCo as dependency. Please setup it following the instructions:
   • Instructions for MuJoCo: https://mujoco.org/

We offer a fine-trained Option Executor (OE) model which Option Planner (OP) utilizes during training for testing.

models/execute.zip

• Training of OE

  • Command

  python train_oe.py

  • The models of oe will be saved at oe_model.
  • The tensorboard logs of oe will be saved at oe_train.

• Training of OP

  • Command

  python train_op.py

  • The models of op will be saved at op_model.
  • The tensorboard logs of op will be saved at op_train.

    @inproceedings{o3f/iros/pang,
      author    = {Jing-Cheng Pang and
                   Si-Hang Yang and
                   Xiong-Hui Chen and
                   Xinyu Yang and
                   Yang Yu and 
                   Mas Ma and
                   Ziqi Guo and
                   Howard Yang and 
                   Bill Huang},
      title     = {Object-Oriented Option Framework for Robotics Manipulation in Clutter},
      booktitle = {{IEEE/RSJ} International Conference on Intelligent Robots and Systems,
                  {IROS}},
      year      = {2023}}