Research map / Reinforcement learning / Visual and representation RL
Diversify \& Conquer: Outcome-directed Curriculum RL via Out-of-Distribution Disagreement
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 D2C instead of prior curriculum methods | Goal-conditioned RL · Complex-Maze, Medium-Maze, Spiral-Maze, Ant Locomotion, Sawyer Push, Sawyer Pick&Place | episode success rate | Helped |
| Use D2C curriculum proposals instead of baseline proposals | Curriculum goal interpolation · Complex-Maze, Medium-Maze, Spiral-Maze, Ant Locomotion, Sawyer Push, Sawyer Pick&Place | distance from curriculum goals to final goals | Helped |
| Replace the proposed intrinsic reward with sparse reward (ablation) | D2C · Sawyer Push | episode success rate | Hurt |
| Remove the curriculum proposal while retaining the intrinsic reward (ablation) | D2C · Spiral-Maze | episode success rate | Hurt |
| Vary the number of conditional-classifier prediction heads (ablation) | D2C · Spiral-Maze and Sawyer Push | episode success rate | No clear effect |
| Vary the auxiliary loss weight λ (ablation) | D2C · Spiral-Maze and Sawyer Push | episode success rate | No clear effect |
| Add value-function bias to the curriculum objective to reflect temporal distance (ablation) | D2C · Complex-Maze, Medium-Maze, Spiral-Maze, Ant Locomotion, Sawyer Push, Sawyer Pick&Place | episode success rate | No clear effect |
| Train conditional classifiers using visited states and desired outcomes as goal candidates rather than the target distribution (ablation) | D2C · Complex-Maze, Medium-Maze, Spiral-Maze, Ant Locomotion, Sawyer Push, Sawyer Pick&Place | episode success rate | No clear effect |
Closest papers by meaning
- Diffusion-based Curriculum Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Outcome-directed Reinforcement Learning by Uncertainty \& Temporal Distance-Aware Curriculum Goal Generation (ICLR 2023 · similarity 0.89)
- Active Curriculum Refinement for Reinforcement Learning (ICML 2026 · similarity 0.89)
- Training Large Language Models for Reasoning through Reverse Curriculum Reinforcement Learning (ICML 2024 · similarity 0.87)
- Learning Progress Driven Multi-Agent Curriculum (ICML 2025 · similarity 0.87)
- AdaCuRL: Adaptive Curriculum Reinforcement Learning with Invalid Sample Mitigation and Historical Revisiting (AAAI 2026 · 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.