Interaction Generation with IPG:

Each agent is solving a distributed trajectory optimization problem using predictions of other agents to avoid collisions. We use the Potential Differential Game formulation to predict the cooperative behavior between agents. We call this an imagined potential game (IPG) since the parameters and cost functions of other agents are not known but imagined by the ego agent. 

We use an iLQR solver to solve the game and a receding horizon control manner to simulate the interaction.

Simulated interactions in different scenarios are shown in the Demo section.