Research map / Reinforcement learning / Visual and representation RL
Decision Pre-Trained Transformer is a Scalable In-Context Reinforcement Learner
Andrei Polubarov, Lyubaykin Nikita, Alexander Derevyagin, Artyom Grishin, Igor Saprygin, Aleksandr Serkov, Mark Averchenko, Daniil Tikhonov, Maksim Zhdanov, Alexander Nikulin, Ilya Zisman, Albina Klepach and 2 more
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 |
|---|---|---|---|
| provide 2500 task-specific transitions (offline) instead of empty context (ablation) | Vintix II · 209 training tasks | normalized score: +4.1% | Improved |
| use flow-based DPT instead of prompted Vintix (AD) on unseen Bi-DexHands | Vintix II · unseen Bi-DexHands | scaled return: +17% | Improved |
| use flow-based DPT instead of prompted Vintix on unseen Meta-World ML45 | Vintix II · unseen Meta-World ML45 | scaled return: +63% | Improved |
| use flow-based DPT instead of prompted Vintix on unseen MuJoCo | Vintix II · unseen MuJoCo | scaled return: +4% | Improved |
| compare with REGENT on Meta-World ML45 (5 tasks) | Vintix II · Meta-World ML45 | normalized return: +8.2% | Improved |
| online (no demos) vs prompted evaluation on unseen tasks (ablation) | Vintix II · 46 unseen tasks | normalized score | Mixed |
| increase demonstration prompt size 500 to 4000 transitions (ablation) | Vintix II · unseen tasks | normalized score | Mixed |
| remove next observation from context tuples (ablation) | Vintix II · not shown | performance | No clear effect |
| increase context length for action-belief analysis (0 to 500 transitions) (ablation) | Vintix II · action beliefs | entropy | Improved |
Closest papers by meaning
- Transformers as Decision Makers: Provable In-Context Reinforcement Learning via Supervised Pretraining (ICLR 2024 · similarity 0.91)
- In-Context Decision Transformer: Reinforcement Learning via Hierarchical Chain-of-Thought (ICML 2024 · similarity 0.90)
- Transformers Provably Implement In-Context Reinforcement Learning with Policy Improvement (NeurIPS 2026 · similarity 0.90)
- Towards Provable Emergence of In-Context Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- In-Context Reinforcement Learning From Suboptimal Historical Data (ICML 2025 · similarity 0.89)
- Transformers Can Learn Temporal Difference Methods for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.89)
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.
