Multi-Agent Generative Adversarial Imitation Learning
About
Imitation learning algorithms can be used to learn a policy from expert demonstrations without access to a reward signal. However, most existing approaches are not applicable in multi-agent settings due to the existence of multiple (Nash) equilibria and non-stationary environments. We propose a new framework for multi-agent imitation learning for general Markov games, where we build upon a generalized notion of inverse reinforcement learning. We further introduce a practical multi-agent actor-critic algorithm with good empirical performance. Our method can be used to imitate complex behaviors in high-dimensional environments with multiple cooperative or competing agents.
Jiaming Song, Hongyu Ren, Dorsa Sadigh, Stefano Ermon• 2018
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Road Crossing | Three-Agent Road Crossing | Collision Rate (per 100 Steps)32 | 16 | |
| Multi-agent Navigation | Two-Agent Swap 100 sampled trajectories | Agent-Agent Collisions97 | 4 | |
| Trajectory Distribution Matching | Two-Agent Swap Robosuite simulation (100 sampled trajectories) | EMD (Agent 1)4.67 | 4 | |
| Trajectory Distribution Matching | Three-Agent Road Crossing | EMD (Agent 1)1.2749 | 4 | |
| Two-Arm Lift | Two-Arm Lift Simulation | Successful Lifts0.00e+0 | 4 |
Showing 5 of 5 rows