1. where it starts

• The simple finite state machine (FSM)
• value iteration

Dynamic Terrain Traversal Skills Using Reinforcement Learning

2. A very important paper

• use motion clips as reference (not only FSM)
• PPO
• Contributions:
  • reference state initialization
  • early stop

DeepMimic: Example-Guided Deep Reinforcement Learning of Physics-Based Character Skills

3. The-state-of-art

• learn multiple skills from unstructured motion clips

AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control

4. my collections (if you are interested)

Procedural Animation