Research map / Reinforcement learning / Visual and representation RL
Learning to Adapt in Dynamic, Real-World Environments through Meta-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 |
|---|---|---|---|
| Meta-train a dynamics model for rapid online adaptation | GrBAL and ReBAL · Simulated continuous-control tasks with changing dynamics | return | Helped |
| Meta-train for fast adaptation rather than applying dynamic evaluation to a standard model | GrBAL and ReBAL vs. MB+DE · Simulated fast-adaptation and generalization tasks | return | Helped |
| Use GrBAL online adaptation | Real-world legged millirobot · Unseen real-world environments and perturbations | trajectory-following performance | Helped |
| Use GrBAL instead of a standard supervised dynamics model | Real-world legged millirobot · Training terrains | trajectory-following cost | No clear effect |
| Adapt model parameters using recent experience (ablation) | GrBAL · Simulated test environments | K-step model prediction error | Helped |
| Train on a larger range of force perturbations (ablation) | GrBAL 7-DOF arm · Force-perturbation test tasks | performance | Helped |
| Train without perturbation forces (ablation) | GrBAL 7-DOF arm · Force-perturbation test tasks | performance | Hurt |
| Vary the adaptation and prediction window lengths K and M (ablation) | GrBAL · Half-cheetah disabled-joint and sloped-terrain tasks | return | No clear effect |
| Use longer sequential inputs with ReBAL rather than gradient-based adaptation with GrBAL | ReBAL vs. GrBAL · Simulated adaptation and generalization tasks | return | Mixed |
| Use model-based meta-RL instead of model-free RL | GrBAL and ReBAL · Simulated continuous-control tasks | meta-training data required for comparable performance: 1000 times less data | Helped |
Closest papers by meaning
- Enabling Adaptive Agent Training in Open-Ended Simulators by Targeting Diversity (NeurIPS 2024 · similarity 0.89)
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.89)
- World Models as Reference Trajectories for Rapid Motor Adaptation (NeurIPS 2025 · similarity 0.88)
- Efficient Test-Time Adaptation For Robot Policies (NeurIPS 2026 · similarity 0.88)
- DynaMITE-RL: A Dynamic Model for Improved Temporal Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- MAMBA: an Effective World Model Approach for Meta-Reinforcement Learning (ICLR 2024 · similarity 0.87)
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.