FlowBelief

Belief-space motion planning experiments using diffusion or flow-matching policies for a custom 2D belief maze environment.

Installation

Create a Python environment, then install the project dependencies:

pip install -r requirements.txt

Project Structure

.
├── belief_env.py          # Custom belief maze environment
├── belief_agent.py        # Belief dynamics and covariance update model
├── belief_API.py          # RRT wrapper and dataset collection entry point
├── Inference.py           # RRT + learned action model inference
├── train.py              # Belief policy training loop
├── train_manager.py      # Configures and launches training
├── cfgs/beliefmaze.yaml  # Training and inference config
├── datasets/             # Belief trajectory dataset
├── metadata/             # Dataset normalization statistics
└── checkpoints/          # Model checkpoints

Usage

Collect belief-space training paths:

python belief_API.py

Train the policy:

python train.py

Run inference:

python Inference.py