Diffused Fields Robotics

This package is supplementary material for the paper "Object-centric Task Representation and Transfer using Diffused Orientation Fields".

This is the robotics package for object-centric robot manipulation applications: local action primitives (peeling, slicing and tactile coverage), trajectory optimization, and reinforcement learning using Diffused Orienation Fields (DOF).

This package depends on the diffused_fields for computing DOF on point clouds.

Installation

Large files (Git LFS required)

This repository uses Git LFS to store large files (e.g. data, models, point clouds).

1. Before cloning

Make sure Git LFS is installed before you clone. You can skip this step if you already have it installed and activated in your machine.

Install Git LFS (Ubuntu)

sudo apt install git-lfs

Install Git LFS (macOS) using homebrew

brew install git-lfs

run once to enable LFS

git lfs install 

This package depends on the diffused_fields library. First clone both repositories:

git clone https://github.com/idiap/diffused_fields.git
git clone https://github.com/idiap/diffused_fields_robotics.git

Create a virtual environment and install both packages in editable mode:

cd diffused_fields_robotics
# Create a virtual environment using Python 3.12
python3.12 -m venv df
# Activate the virtual environment
source df/bin/activate
# Install diffused_fields in editable mode using path to its root directory
pip install -e ../diffused_fields
# Install diffused_fields_robotics in editable mode
pip install -e .

Paper and Citation

If you use this package in your research, please cite: (Coming soon)

Quick Start

Running Action Primitives

# Run slicing on a banana
python scripts/slicing.py

# Run peeling on a pear
python scripts/peeling.py

# Run coverage on a surface
python scripts/coverage.py

Running Batch Experiments

See scripts/batch_experiments/README.md for details.

Transfer across objects:

python scripts/batch_experiments/batch_peeling.py
python scripts/analysis/batch_peeling_stats_primitives.py

Robustness to noise:

python scripts/batch_experiments/batch_slicing_geometric_noise.py
python scripts/analysis/robustness.py

Reproducing Paper Results

All simulation data and plots from the paper can be generated using the scripts in scripts/batch_experiments/ and scripts/analysis/.

---

This code is maintained by Cem Bilaloglu and licensed under the MIT License.

Copyright (c) 2025 Idiap Research Institute - [email protected]