- Solve the issue about avail_actions;
- Improve some algorithms, such as SAC, MASAC, DDPG, and TD3, etc.
- Add multi-robot warehouse (RWARE) environment.
- n_size -> buffer_size;
- The saving and loading of observation status;
- Unify the names of attributes for both single-agent and multi-agent DRL.
- Support MetaDrive environment, and provide examples;
- Refine the configuration settings for the gym-pybullet-drones environment;
- Fix some issues about models saving;
- Revise the model loading method;
- Implement a configuration option for selecting different activation function of the output layer of actors for continuous control;
- Update the corresponding examples to adapt the above changes;
- Fix some other bugs.
- Support and finish test for gym-pybullet-drones environments;
- Fix some issues for installation of xuance. Now it is more easy to install and use;
- Improve the compatibility for MacOS with Apple's M chips;
- Fix some other bugs.
View at https://pypi.org/project/xuance/1.0.11/.
- Added MiniGrid environment with PPO implementations;
- Added gym-pybullet-drones UAV environment with PPO implementation;
- Fixed the issue of failed parameter reading for multi-agent reinforcement learning with adversarial tasks;
- Fixed several bugs under the TensorFlow and MindSpore frameworks.