Formal Contracts for Multi-Agent Reinforcement Learning (contracts library)

The Formal Contracts for Multi-Agent Reinforcement Learning project provides an RLlib-based codebase for specifying and enforcing contract structures between agents so that individually learning agents achieve cooperative outcomes in social dilemma settings, accompanying the paper demonstrating how such formal contracts can improve cooperation in multi-agent RL benchmarks.

Endorsements support MIT Algorithmic Alignment Group.

No endorsements yet.

Discussion

Start Date

-

End Date

-

Expected Duration

-

Funding Raised to Date

-