Habitizing Diffusion Planning for Efficient and Effective Decision Making

May 26, 2025 · View on GitHub

· Paper · Code · Webpage ·

This repository contains the PyTorch implementation of "Habitizing Diffusion Planning for Efficient and Effective Decision Making." (ICML 2025)



🛠️ Setup

Prerequisites

It is recommended to set up a dedicated conda environment.

Create a Conda Environment

conda create -n Habi python=3.9 mesalib glew glfw pip=23 setuptools=63.2.0 wheel=0.38.4 protobuf=3.20 -c conda-forge -y
conda activate Habi

Install MuJoCo Simulator and mujoco-py (Important)

Follow the official installation guide here. Alternatively, you can use the script below:

#!/bin/bash
sudo apt-get update && sudo apt-get install -y wget tar libosmesa6-dev libgl1-mesa-glx libglfw3 patchelf cmake
sudo ln -s /usr/lib/x86_64-linux-gnu/libGL.so.1 /usr/lib/x86_64-linux-gnu/libGL.so

# Set up MuJoCo
USER_DIR=$HOME
wget -c "https://mujoco.org/download/mujoco210-linux-x86_64.tar.gz"
mkdir -p $USER_DIR/.mujoco
cp mujoco210-linux-x86_64.tar.gz $USER_DIR/mujoco.tar.gz
rm mujoco210-linux-x86_64.tar.gz
tar -zxvf $USER_DIR/mujoco.tar.gz -C $USER_DIR/.mujoco

# Update environment variables
echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$USER_DIR/.mujoco/mujoco210/bin" >> ~/.bashrc
echo "export MUJOCO_PY_MUJOCO_PATH=$USER_DIR/.mujoco/mujoco210" >> ~/.bashrc
source ~/.bashrc

Install Dependencies

pip install -r requirements.txt
pip install -e .

Additionally, you may need to install PyTorch following the official setup guide depending on your system configuration.

💻 Training

Pretraining Habitization Target

To pretrain the habitization target, run:

bash pretrain_habi_target.sh

This script provide training for example diffusion planners. You can modifiy the pipelines/planner_d4rl_TASK.py to integrate any diffusion planner of your choice.

Training Habi and Evaluating Habi

To train the Habi model across different tasks and seeds, run:

bash run_habi.sh

This script trains habi across various Mujoco, Kitchen, AntMaze, and Maze2D tasks with multiple random seeds for robustness. This script will transfer the diffusion planner into the corresponding habitual behavior policy, enabling fast inference and deployment.

📌 Evaluation Tasks

We evaluate Habi across multiple environments, across different types of decision-making tasks, including locomotion, manipulation, and navigation:

  • MuJoCo:

    • HalfCheetah-Medium-Expert-v2
    • HalfCheetah-Medium-Replay-v2
    • HalfCheetah-Medium-v2
    • Hopper-Medium-Expert-v2
    • Hopper-Medium-Replay-v2
    • Hopper-Medium-v2
    • Walker2d-Medium-Expert-v2
    • Walker2d-Medium-Replay-v2
    • Walker2d-Medium-v2

  • Kitchen:

    • Kitchen-Partial-v0
    • Kitchen-Mixed-v0

  • AntMaze:

    • AntMaze-Medium-Play-v2
    • AntMaze-Medium-Diverse-v2
    • AntMaze-Large-Play-v2
    • AntMaze-Large-Diverse-v2

  • Maze2D:

    • Maze2d-Umaze-v1
    • Maze2d-Medium-v1
    • Maze2d-Large-v1

These environments follow the D4RL benchmark.

🏷️ Acknowledgements

This code is built upon the Cleandiffuser repo for consistent evaluation. Please see the license for further details.