Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Maximum Entropy Population-Based Training for Zero-Shot Human-AI Coordination

About

We study the problem of training a Reinforcement Learning (RL) agent that is collaborative with humans without using any human data. Although such agents can be obtained through self-play training, they can suffer significantly from distributional shift when paired with unencountered partners, such as humans. To mitigate this distributional shift, we propose Maximum Entropy Population-based training (MEP). In MEP, agents in the population are trained with our derived Population Entropy bonus to promote both pairwise diversity between agents and individual diversity of agents themselves, and a common best agent is trained by paring with agents in this diversified population via prioritized sampling. The prioritization is dynamically adjusted based on the training progress. We demonstrate the effectiveness of our method MEP, with comparison to Self-Play PPO (SP), Population-Based Training (PBT), Trajectory Diversity (TrajeDi), and Fictitious Co-Play (FCP) in the Overcooked game environment, with partners being human proxy models and real humans. A supplementary video showing experimental results is available at https://youtu.be/Xh-FKD0AAKE.

Rui Zhao, Jinming Song, Yufeng Yuan, Hu Haifeng, Yang Gao, Yi Wu, Zhongqian Sun, Yang Wei• 2021

Related benchmarks

TaskDatasetResultRank
Cooperative Multi-Agent CoordinationOvercooked-AI Asymmetric Advantages
Mean Reward177.5
22
Zero-shot CoordinationOvercooked-AI Coordination Ring
Mean Team Reward152
18
Zero-shot CoordinationOvercooked-AI Cramped Room
Mean Team Reward168.8
18
Zero-shot CoordinationOvercooked-AI Counter Circuit
Mean Team Reward65
18
Zero-shot CoordinationOvercooked-AI Forced Coordination
Mean Team Reward35
18
CoordinationOvercooked Cramped Room layout v1
SP203
14
Multi-agent coordinationOvercooked Coord. Ring
Average Return40.3
10
Human-Agent CoordinationOvercooked Counter Circuit (human evaluation)
Average Score76.19
9
Human-Agent CoordinationOvercooked Multi-strategy Counter (human evaluation)
Average Score64.02
9
Multi-agent coordinationOvercooked Asymm. Adv.
Return127.4
8
Showing 10 of 25 rows

Other info

Follow for update