Research map / Reinforcement learning / Visual and representation RL
Evolving Curricula with Regret-Based Environment Design
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 |
|---|---|---|---|
| evolve edited high-regret levels rather than relying on random level sampling alone | ACCEL · MiniGrid zero-shot mazes, Mean | success rate: 0.71 ± 0.04 -> 0.88 ± 0.04 | Helped |
| evolve edited high-regret levels rather than relying on random level sampling alone | ACCEL · 51×51 procedurally-generated maze | success rate: 25% -> 53% | Helped |
| use an empty starting generator and edit levels as complexity grows | ACCEL · 101×101 procedurally-generated maze | mean success rate: 4% -> 8% | Helped |
| initialize levels in the easy range before editing | ACCEL · BipedalWalker test environments, Mean | return: 102.5 ± 13.0 -> 202.8 ± 13.6 | Helped |
| add level editing to the PLR curriculum (ablation) | MiniGrid curriculum · MiniGrid zero-shot transfer | performance | Helped |
| add level editing to the PLR curriculum (ablation) | BipedalWalker curriculum · BipedalWalker | performance | Helped |
| start from simple levels rather than DR-sampled levels (ablation) | ACCEL · BipedalWalker | performance | Helped |
| replace random edits with an RL-trained learned editor (ablation) | ACCEL · MiniGrid zero-shot transfer | success rate | Mixed |
| replace level editing with additional DR sampling (ablation) | ACCEL · MiniGrid zero-shot transfer | success rate | Mixed |
| replace editing with additional random DR sampling (ablation) | ACCEL · MiniGrid 16Rooms | success rate: 1.0 ± 0.0 -> 0.84 ± 0.06 | Hurt |
| replace editing with additional random DR sampling (ablation) | ACCEL · MiniGrid SimpleCrossing | success rate: 0.8 ± 0.05 -> 0.75 ± 0.05 | Hurt |
| replace editing with additional random DR sampling (ablation) | ACCEL · MiniGrid 16Rooms2 | success rate: 0.51 ± 0.28 -> 0.68 ± 0.18 | Helped |
Closest papers by meaning
- No Regrets: Investigating and Improving Regret Approximations for Curriculum Discovery (NeurIPS 2024 · similarity 0.92)
- Marginal Benefit Driven RL Teacher for Unsupervised Environment Design (AAAI 2025 · similarity 0.90)
- Improving Regret Approximation for Unsupervised Dynamic Environment Generation (NeurIPS 2025 · similarity 0.90)
- Reward-Free Curricula for Training Robust World Models (ICLR 2024 · similarity 0.90)
- Improving Environment Novelty Quantification for Effective Unsupervised Environment Design (NeurIPS 2024 · similarity 0.90)
- Refining Minimax Regret for Unsupervised Environment Design (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.