R³ Legacy Release
This release is a snapshot of the original RTSS 2023 paper implementation of R³: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics.
📄 Paper: R³: On-device Real-Time Deep Reinforcement Learning for Autonomous Robotics (RTSS 2023)
Dependency Stack (frozen)
| Component | Version |
|---|---|
| Python | 3.8.10 |
| PyTorch | 1.13.0 / 1.13.1 |
| Tensorflow | 1.15.5+nv22.12 |
| gym | 0.15.3 |
| gym-donkeycar | 1.3.0 |
| autonomous-learning-library | 0.9.x (vendored) |
| JetPack (Jetson) | 5.0.2 |
| OS (Jetson) | Ubuntu 20.04.5 LTS |
| Hardware tested | NVIDIA Jetson AGX Orin / Xavier |
Recommended use
- ✅ Use this release if you want to reproduce the exact numbers from the RTSS 2023 paper.
- ✅ Use this release if you are still on JetPack 5.0.2 and the legacy CUDA / cuDNN stack.
⚠️ For new deployments on JetPack 6.x with PyTorch 2.x and Gymnasium, see the upcomingupgrade/pytorch-2.5-gymnasiumbranch.
Installation
Follow the original README instructions in this tag. Highlights:
conda create -n rapidlearn python=3.8.10
conda activate rapidlearn
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
cd gym-donkeycar && python setup.py install
cd ../MUSHR-DL && pip install -r requirements.txtCitation
If you use this work, please cite the RTSS 2023 paper.
This is the legacy/baseline snapshot. Future modernization work (PyTorch 2.5 / Gymnasium / Python 3.10 / JetPack 6.2) will live on a separate branch and will be released as v2.x.