TJU-DRL-LAB/AI-Optimizer

Covers model-free and model-based approaches alongside multi-agent, offline, and self-supervised variants with specialized algorithm libraries. Built on a modular distributed training framework supporting permutation-invariant network architectures for scalability challenges like dimensional explosion and non-stationarity in cooperative settings. Integrates with standard benchmarks including SMAC and particle environments, with implementations like API-QMIX achieving near-perfect win rates on hard exploration scenarios.

3,462 stars. No commits in the last 6 months.