Research map / Reinforcement learning / Visual and representation RL
Off-Team Learning
Brandon Cui, Hengyuan Hu, Andrei Lupu, Samuel Sokota, Jakob Nicolaus Foerster
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 |
|---|---|---|---|
| Train belief model off-team (OT-BL) instead of vanilla belief learning | OBL · 2-life Hanabi, level 1, with invalid card penalty | cross-play score: 16.71 ± 0.06 -> 21.15 ± 0.02 | Improved |
| OT-BL instead of vanilla belief learning | OBL · 2-life Hanabi, level 3, no penalty | cross-play score: 23.05 ± 0.30 -> 23.76 ± 0.02 | Improved |
| OT-BL instead of vanilla belief learning | OBL · 2-life Hanabi, level 1, no penalty | SXG: 0.84 -> 0.05 | Improved |
| OT-BL instead of vanilla belief learning | belief model · 2-life Hanabi with penalty, level 1 policy training | % valid fictitious samples: ~83% -> > 99.5% | Improved |
| Add invalid card penalty to environment | vanilla OBL level 1 · 2-life Hanabi | cross-play score: 20.17 ± 0.15 -> 16.71 ± 0.06 | Hurt |
| Add invalid card penalty to environment | OT-BL level 1 · 2-life Hanabi | cross-play score: 21.05 ± 0.02 -> 21.15 ± 0.02 | No clear effect |
| OT-OBL instead of vanilla OBL | OBL · 3-life Hanabi, level 5 | cross-play score: 23.99 ± 0.11 -> 24.24 ± 0.01 | Improved |
| OT-OBL at level 1 | OBL · 3-life Hanabi, level 1 | cross-play score: 21.50 ± 0.05 -> 21.31 ± 0.01 | Hurt |
| Add robust belief (belief trained on broader distributions) to OT-OBL | OT-OBL · 3-life Hanabi, level 5 | cross-play score: 24.24 ± 0.01 -> 24.30 ± 0.01 | Improved |
| OT-OBL with 3200 rank bot trajectories | OBL · ad-hoc teamplay with rank bot | score: 14.47 ± 0.67 -> 15.66 ± 0.56 | Improved |
| OT-OBL with 3200 clone bot trajectories | OBL · proxy human-AI coordination | score: 16.19 ± 0.21 -> 16.64 ± 0.20 | Improved |
| OT-OBL with clone bot trajectories evaluated against rank bot | OBL · ad-hoc teamplay with rank bot | score: 14.47 ± 0.67 -> 14.33 ± 0.63 | No clear effect |
Closest papers by meaning
- Off-Belief Learning (ICML 2021 · similarity 0.92)
- Open Ad Hoc Teamwork with Cooperative Game Theory (ICML 2024 · similarity 0.85)
- OvercookedV2: Rethinking Overcooked for Zero-Shot Coordination (ICLR 2025 · similarity 0.85)
- Differentiable Belief-based Opponent Shaping (NeurIPS 2026 · similarity 0.85)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.84)
- ODRL: A Benchmark for Off-Dynamics Reinforcement Learning (NeurIPS 2024 · similarity 0.84)
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.
