v2.5.0: New algorithm (SimBa in SBX) and NumPy 2.0 support

SB3 Contrib (more algorithms): https://github.com/Stable-Baselines-Team/stable-baselines3-contrib
RL Zoo3 (training framework): https://github.com/DLR-RM/rl-baselines3-zoo
Stable-Baselines Jax (SBX): https://github.com/araffin/sbx

To upgrade:

pip install stable_baselines3 sb3_contrib rl_zoo3 --upgrade

Breaking Changes:

• Increased minimum required version of PyTorch to 2.3.0
• Removed support for Python 3.8

New Features:

• Added support for NumPy v2.0: VecNormalize now cast normalized rewards to float32, updated bit flipping env to avoid overflow issues too
• Added official support for Python 3.12

SBX (SB3 + Jax)

• Added SimBa Policy: Simplicity Bias for Scaling Up Parameters in DRL
• Added support for parameter resets

Others:

• Updated Dockerfile

Documentation:

• Added Decisions and Dragons to resources. (@jmacglashan)
• Updated PyBullet example, now compatible with Gymnasium
• Added link to policies for policy_kwargs parameter (@kplers)
• Add FootstepNet Envs to the project page (@cgaspard3333)
• Added FRASA to the project page (@MarcDcls)
• Fixed atari example (@chrisgao99)
• Add a note about Discrete action spaces with start!=0
• Update doc for massively parallel simulators (Isaac Lab, Brax, ...)
• Add dm_control example

New Contributors

• @jmacglashan made their first contribution in #2044
• @kplers made their first contribution in #2050
• @MarcDcls made their first contribution in #2059
• @cgaspard3333 made their first contribution in #2058
• @sanowl made their first contribution in #2064
• @chrisgao99 made their first contribution in #2071

Full Changelog: v2.4.0...v2.5.0