Featured Use Cases
Emergent Collaborative and Competitive Behavior
In this experiment, we study how collaborative and competitive behaviors emerge among agents in a partially observable stochastic game. In our simulation, each agent occupies a square and can move around the map. Each agent can “attack” agents that are on a different “team”; the attacked agent loses its life and is removed from the simulation. Each agent can observe the state of the map in a region surrounding its location. It can see other agents and what team they’re on as well as the edges of the map. The diagram below visuially depicts the agents’ observation and action spaces.
AI-Enabled Conflict Simulation
We use Abmarl’s simulation interface to connect a C++ based conflict simulation JCATS to reinforcement learning algorithms in order to train an agent to navigate to a waypoint. All state updates are controlled by the JCATS simulation itself. Positional observations are reported to the RL policy, which in turn issues movement commands to the the simulator. We leveraged Abmarl as a proxy simulation to rapidly find a warm start configuration. Training is performed on a cluster of 4 nodes utilizing RLlib’s client-server architecture. We successfully generated 136 million training steps and trained the agent to navigate the scenario.