Research map / Multi-agent systems and games / Cooperative multi-agent RL
PADiff: Predictive and Adaptive Diffusion Policies for Ad Hoc Teamwork
Hohei Chan, Xinzhi Zhang, Antao Xiang, Weinan Zhang, Mengchen Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace the MLP denoiser with AFM-Net (ablation) | PADiff · PP, LBF, Overcooked | average return | Improved |
| replace the U-Net denoiser with AFM-Net (ablation) | PADiff · PP, LBF, Overcooked | average return | Improved |
| add the CoGoal predictor in the Predictive Guidance Block (ablation) | PADiff · PP, LBF, Overcooked | average return | Improved |
| add the CoReturn predictor in the Predictive Guidance Block (ablation) | PADiff · PP, LBF, Overcooked | average return | Improved |
| increase diffusion steps beyond 30 (ablation) | PADiff · PP, LBF, Overcooked | average return | No clear effect |
| use a moderate dropout rate with suitably chosen diffusion depth (2, 10, 20, 30 steps) (ablation) | PADiff · PP, LBF, Overcooked | average return | Improved |
| PADiff vs strongest baseline TAGET | PADiff · Predator-Prey (4 test policies) | average return: 61.2 -> 67.0 (+9.47%) | Improved |
| PADiff vs strongest baseline TAGET | PADiff · LBF (4 test policies) | average return: 0.080 -> 0.130 (+62.50%) | Improved |
| PADiff vs strongest baseline CQL | PADiff · Overcooked (4 test policies) | average return: 0.64 -> 0.88 (+37.50%) | Improved |
| PADiff vs strongest baseline TAGET | PADiff · Predator-Prey (8 test policies) | average return: 60.1 -> 65.7 (+9.32%) | Improved |
| PADiff vs strongest baseline LIAM | PADiff · LBF (8 test policies) | average return: 0.065 -> 0.117 (+80.00%) | Improved |
| PADiff vs strongest baseline TAGET | PADiff · Overcooked (8 test policies) | average return: 0.63 -> 0.71 (+12.70%) | Improved |
Closest papers by meaning
- ROTATE: Regret-driven Open-ended Training for Ad Hoc Teamwork (NeurIPS 2026 · similarity 0.89)
- Ad Hoc Teamwork via Offline Goal-Based Decision Transformers (ICML 2025 · similarity 0.88)
- N-agent Ad Hoc Teamwork (NeurIPS 2024 · similarity 0.88)
- MADiff: Offline Multi-agent Learning with Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Minimum Coverage Sets for Training Robust Ad Hoc Teamwork Agents (AAAI 2024 · similarity 0.86)
- NestRL: A Nested Training Regime for Mutual Adaptation in Human–AI Teaming (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
