Multi-Agent Pursuit and Evasion
A team of agents pursues an evader who moves arbitrarily. The pursuers seek to capture the evader without collision with teammates and environment, while the evader trying to escape from the pursuers.
Features and Methods
- No peer communication and dynamics models
- Decentralized decision making.
- Agents learn through interactions with other agents and environment.
- Reinforment learning
- Captured one evader whose motion was pre-defined.
- Agents learned achieving goal with consideration of safety constraints.
- Learned control policy is robust against uncertainties.