API Change

1. Expand the Python version support for DI-engine to Python3.7-Python3.10

Env

1. add pistonball MARL env and its unittest/example (#833)
2. update trading env (#831)
3. update ppo config for better discrete action space performance (#809)
4. remove unused config fields in MuJoCo PPO

Algorithm

1. add AWR algorithm (#828)
2. add encoder in MAVAC (#823)
3. add HPT model architecture (#841)
4. fix multiple model wrappers reset bug (#846)
5. add hybrid action space support to ActionNoiseWrapper (#829)
6. fix mappo adv compute bug (#812)

Enhancement

1. add resume_training option to allow the envstep and train_iter resume seamlessly (#835)
2. polish old/new pipeline DistributedDataParallel (DDP) implementation (#842)
3. adapt DingEnvWrapper to gymnasium (#817)

Fix

1. fix priority buffer delete bug (#844)
2. fix middleware collector env reset bug (#845)
3. fix many unittest bugs

Style

1. downgrade pyecharts log level to warning and polish installation doc (#838)
2. polish necessary requirements
3. polish api doc details
4. polish DI-engine citation authors
5. upgrade CI macos version from 12 to 13

News

1. CleanS2S: High-quality and streaming Speech-to-Speech interactive agent in a single file.
2. GenerativeRL: Revisiting Generative Policies: A Simpler Reinforcement Learning Algorithmic Perspective
3. PRG: Pretrained Reversible Generation as Unsupervised Visual Representation Learning

Full Changelog: https://github.com/opendilab/DI-engine/compare/v0.5.2...v0.5.3

Contributors: @PaParaZz1 @puyuan1996 @kxzxvbk @YinminZhang @zjowowen @luodi-7 @MarkHolmstrom @TairanMK