v0.4.0

Release notes

Enhancements

• Easy Installation: GPUDrive is now available as an easy-to-install package, with both manual and Docker installation options.
• Improved data loading and iteration: Train on many scenes in the WOMD with new data loaders for improved data handling.
• Improved PPO Loop: Optimized the PPO loop, leading to better convergence and improved baseline performance.
• Pre-trained Policy: Released a pre-trained policy, achieving 99% performance in the simulator.
• Visualization enhancements: Upgraded 2D and 3D visualizers for improved rendering quality and a more intuitive user experience.

Other updates

• ICLR 2025 Acceptance: GPUDrive has been accepted to ICLR 2025! The updated paper is now available here

Documentation

• Updated tutorials and readme: All tutorials and documentation have been tested and updated to reflect the latest improvements.

v0.3.0

New version of the WOMD processed dataset with GPUDrive. Warning: Breaking changes from the previous Nocturne dataset. Use the updated datasets linked in the README. It also contains a new visualizer; using matplotlib instead of pygame.

v0.2.0

All RL code trains agents successfully. Some quality of life fixes are missing.

v0.1.0-alpha

Readme (#183)

* conda env

* Update docs

* Make more changes

* Add a gif

* Add a gif

* Add gifs

* Remove global gif

* Udpated