1. Reproducible

We provide algorithms that reproduce stably the results of many influential reinforcement learning algorithms.

2. Large Scale

Ability to support high-performance parallelization of training with thousands of CPUs and multi-GPUs.

3. Reusable

Algorithms provided in the repository can be directly adapted to new tasks by defining a forward network and training mechanism will be built automatically.

4. Extensible

Build new algorithms quickly by inheriting the abstract class in the framework.