Research map / Reinforcement learning / Visual and representation RL
Provably Efficient Model-based Policy Adaptation
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 |
|---|---|---|---|
| Use model-based policy adaptation with aggregated target-environment data | PADA-DM on continuous control tasks · Continuous control tasks | training samples to convergence: 10k to 50k training samples; [Christiano et al., 2016] requires 5 times more samples than our methods on average | Helped |
| Use PADA-DM rather than model-free policy optimization | PADA-DM versus PPO · Ant | episodic reward | Helped |
| Use PPO with target-environment reward optimization | PPO compared with PADA methods · HalfCheetah and Reacher | episodic reward | Mixed |
| Adapt online using PADA-DM in addition to domain randomization (ablation) | Domain-randomized source policy · Continuous control tasks | episodic reward | Helped |
| Use domain randomization without online target adaptation (ablation) | Source policy · Target domains outside the domain-randomization distribution | episodic reward | No clear effect |
| Replace MPC action selection with a learned target policy (ablation) | PADA-DM · Continuous control tasks | episodic reward | No clear effect |
| Optimize CEM using target-environment reward instead of state deviation (ablation) | PADA-DM · Target-environment policy adaptation | episodic reward | Hurt |
| Use a nonlinear rather than linear deviation model (ablation) | Deviation model · Deviation model prediction | prediction accuracy | Helped |
| Use the learned source dynamics model instead of ground-truth source dynamics (ablation) | PADA-DM · Continuous control tasks | episodic reward | No clear effect |
| Adapt in a target environment with high action-noise divergence | PADA methods · Ant-v2 0.6 std motor noise | episodic reward | Hurt |
Closest papers by meaning
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.89)
- Learning to Reuse Policies in State Evolvable Environments (ICML 2025 · similarity 0.89)
- Reinforcement Learning with Adaptive Regularization for Safe Control of Critical Systems (NeurIPS 2024 · similarity 0.88)
- From Static Policies to Adaptive Priors in Offline Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.88)
- Pre-Trained Multi-Goal Transformers with Prompt Optimization for Efficient Online Adaptation (NeurIPS 2024 · similarity 0.88)
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.