Research map / Reinforcement learning / Visual and representation RL
Self-Correcting Models for Model-Based Reinforcement Learning
Erik Talvitie
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 |
|---|---|---|---|
| Use hallucinated training instead of standard DAgger-MC training | MBRL with a flawed model on Shooter with moving bullseyes · Shooter, moving bullseyes | discounted return | Improved |
| Use hallucinated training instead of standard DAgger-MC training | MBRL when the model class contains a perfect model · Shooter, fixed bullseyes | discounted return | Hurt |
| Use hallucinated one-step error as a training objective in stochastic environments | Model-based reinforcement learning · Stochastic MDPs | validity as a proxy for multi-step prediction error | Hurt |
| Use hallucinated one-step error with a perfect model and a stochastic policy | Model-based reinforcement learning · Deterministic MDPs with arbitrary policies | hallucinated one-step error | Hurt |
| Use separate models for each rollout depth instead of one shared model (ablation) | H-DAgger-MC · Shooter, fixed bullseyes | discounted return | Improved |
| Use a single model across rollout depths instead of separate depth-specific models (ablation) | H-DAgger-MC · Shooter, fixed bullseyes, training rollouts to depth 15 | discounted return | Hurt |
| Truncate training rollouts at shorter depths (ablation) | H-DAgger-MC with a single shared model · Shooter, fixed bullseyes | discounted return | Improved |
| Train on deep hallucinated rollouts from the beginning rather than gradually increasing rollout depth (ablation) | H-DAgger-MC · Shooter | training effectiveness | Hurt |
Closest papers by meaning
- Plan To Predict: Learning an Uncertainty-Foreseeing Model For Model-Based Reinforcement Learning (NeurIPS 2022 · similarity 0.86)
- On Rollouts in Model-Based Reinforcement Learning (ICLR 2025 · similarity 0.85)
- Policy-shaped prediction: avoiding distractions in model-based reinforcement learning (NeurIPS 2024 · similarity 0.85)
- Trust the Model Where It Trusts Itself - Model-Based Actor-Critic with Uncertainty-Aware Rollout Adaption (ICML 2024 · similarity 0.83)
- BECAUSE: Bilinear Causal Representation for Generalizable Offline Model-based Reinforcement Learning (NeurIPS 2024 · similarity 0.82)
- Robust Model Based Reinforcement Learning Using $\mathcal{L}_1$ Adaptive Control (ICLR 2024 · similarity 0.82)
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.
