Research map / Reinforcement learning / Offline RL
Bootstrapped Transformer for Offline Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| bootstrap Trajectory Transformer with self-generated trajectories | Trajectory Transformer · D4RL Gym average | normalized score: 70.1 -> 81.9 | Helped |
| bootstrap Trajectory Transformer on Adroit | Trajectory Transformer · D4RL Adroit average | normalized score: 20.1 -> 23.7 | Helped |
| retrain TT on original data for same steps (control for extra training) (ablation) | Trajectory Transformer · D4RL Gym average | normalized score: 70.1 -> 69.4 | No clear effect |
| add Gaussian noise to trajectories (S4RL) instead of model-generated data (ablation) | Trajectory Transformer · D4RL Gym average | normalized score: 70.1 -> 71.1 (All), 70.9 (Last) | No clear effect |
| autoregressive instead of teacher-forcing generation (bootstrap once) (ablation) | BooT-o · D4RL Gym average | normalized score: 79.9 -> 81.9 | Helped |
| bootstrap repeatedly (BooT-r) instead of once (BooT-o), autoregressive (ablation) | BooT AR · D4RL Gym average | normalized score: 81.9 -> 79.1 | Hurt |
| bootstrap repeatedly instead of once, teacher forcing (ablation) | BooT TF · D4RL Gym average | normalized score: 79.9 -> 80.9 | Helped |
| start bootstrapping later (increase threshold k) (ablation) | BooT · D4RL Gym | normalized score | Mixed |
| increase generation percentage eta (ablation) | BooT · D4RL Gym | normalized score | Mixed |
| increase generation length T' (ablation) | BooT TF · D4RL Gym | normalized score | Hurt |
| retrain TT on lowest-loss data instead of random (ablation) | TT-retrain · D4RL Gym | normalized score | Helped |
| model-generated data instead of random noise for augmentation | BooT-o AR · Generated vs original dataset | RMSE: 21.5 -> 16.2 | Helped |
Closest papers by meaning
- Tackling Data Corruption in Offline Reinforcement Learning via Sequence Modeling (ICLR 2025 · similarity 0.89)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.89)
- When should we prefer Decision Transformers for Offline Reinforcement Learning? (ICLR 2024 · similarity 0.89)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Stitching Sub-trajectories with Conditional Diffusion Model for Goal-Conditioned Offline RL (AAAI 2024 · similarity 0.89)
- Q-value Regularized Transformer for Offline Reinforcement Learning (ICML 2024 · 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.