On the Robustness of Cooperative Multi-Agent Reinforcement Learning
About
In cooperative multi-agent reinforcement learning (c-MARL), agents learn to cooperatively take actions as a team to maximize a total team reward. We analyze the robustness of c-MARL to adversaries capable of attacking one of the agents on a team. Through the ability to manipulate this agent's observations, the adversary seeks to decrease the total team reward. Attacking c-MARL is challenging for three reasons: first, it is difficult to estimate team rewards or how they are impacted by an agent mispredicting; second, models are non-differentiable; and third, the feature space is low-dimensional. Thus, we introduce a novel attack. The attacker first trains a policy network with reinforcement learning to find a wrong action it should encourage the victim agent to take. Then, the adversary uses targeted adversarial examples to force the victim to take this action. Our results on the StartCraft II multi-agent benchmark demonstrate that c-MARL teams are highly vulnerable to perturbations applied to one of their agent's observations. By attacking a single agent, our attack method has highly negative impact on the overall team reward, reducing it from 20 to 9.4. This results in the team's winning rate to go down from 98.9% to 0%.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Adversarial Attack | SMAC 1c3s5z | Reward14.8 | 12 | |
| Adversarial Attack | MPE spread | Reward Score-953.3 | 12 | |
| Adversarial Attack | SMAC 8m | Reward15.08 | 12 | |
| Adversarial Attack | SMAC bane_vs_bane | Reward16.02 | 12 | |
| Adversarial Attack | SMAC 27m_vs_30m | Reward16.78 | 12 | |
| Adversarial Attack | Google Research Football counterattack | Reward1.27 | 12 | |
| Adversarial Attack | Google Research Football 3 vs 1 | Reward1.75 | 12 | |
| Adversarial Attack | Multi-Agent Particle Environment reference | Reward-33.74 | 12 | |
| Attack Detection | SMAC 1c3s5z | F1 Score66 | 5 | |
| Attack Detection | SMAC 8m | F1 Score69 | 5 |