Research map / Reinforcement learning / MDPs and planning
A Meta-MDP Approach to Exploration for Lifelong Reinforcement Learning
Francisco Garcia, Philip S. Thomas
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 |
|---|---|---|---|
| add learned advisor exploration policy to REINFORCE | REINFORCE · Pole-balance (d) | average return: 20.32 -> 28.52 | Improved |
| add learned advisor exploration policy to REINFORCE | REINFORCE · Animat | average return: -779.62 -> -387.27 | Improved |
| add advisor to PPO | PPO · Pole-balance (d) | average return: 27.87 -> 46.29 | Improved |
| add advisor to PPO | PPO · Animat | average return: -751.40 -> -631.97 | Improved |
| add advisor to PPO | PPO · Pole-balance (c) | average return: 29.95 -> 438.13 | Improved |
| add advisor to PPO | PPO · Hopper | average return: 13.82 -> 164.43 | Improved |
| add advisor to PPO | PPO · Ant | average return: -42.75 -> 83.76 | Improved |
| PPO+Advisor versus MAML | PPO+Advisor · Ant | average return: 113.33 (MAML) vs 83.76 | Hurt |
| use advisor for fine-tuning to new cars | Self-driving agent · Self-driving fine-tuning | episodes to threshold: 42 vs 12 (first half); 5 vs 18 (second half) | Mixed |
| use advisor exploration policy alone as exploitation policy instead of task-specific policy (ablation) | Advisor policy · Pole-balancing and animat unseen tasks | return / steps to goal | Hurt |
Closest papers by meaning
- Discovering Temporally-Aware Reinforcement Learning Algorithms (ICLR 2024 · similarity 0.88)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.88)
- Efficient Preference-Based Reinforcement Learning: Randomized Exploration meets Experimental Design (NeurIPS 2025 · similarity 0.88)
- Can Learned Optimization Make Reinforcement Learning Less Difficult? (NeurIPS 2024 · similarity 0.87)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.87)
- Beyond Optimism: Exploration With Partially Observable Rewards (NeurIPS 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.
