Research map / Reinforcement learning / Visual and representation RL
An Investigation of Model-Free Planning
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 DRC(3,3) instead of VIN | Gridworld 32x32 · Gridworld 32x32, 1e7 steps | % solved: 97 -> 99 | Helped |
| Use DRC(3,3) instead of VIN | Gridworld 32x32 · Gridworld 32x32, 1e6 steps | % solved: 80 -> 30 | Hurt |
| Use DRC(3,3) instead of the strongest listed baseline | Sokoban unfiltered · Sokoban unfiltered test, 2e7 steps | % solved: 25 -> 80 | Helped |
| Use DRC(3,3) instead of the strongest listed baseline | Sokoban unfiltered · Sokoban unfiltered test, 1e9 steps | % solved: 96 -> 99 | Helped |
| Add extra no-op computation steps at episode start | DRC networks · Sokoban medium-difficulty levels | % solved: about 5% | Helped |
| Add extra no-op computation steps at episode start | LSTM(1,1) · Sokoban medium-difficulty levels | % solved | No clear effect |
| Use DRC(3,3) instead of DRC(1,1) | Sokoban medium test, models trained on Medium-size dataset · Sokoban | generalization gap: 33.5% -> 6.5% | Helped |
| Use DRC(3,3) instead of DRC(1,1) | Sokoban, extrapolating from 4 to 7 boxes · Sokoban 7-box levels | performance degradation: 18.5% -> 3.5% | Helped |
| Use nine stacked modules without repeats instead of DRC(3,3) (ablation) | Sokoban medium · Sokoban medium test, 1e9 steps | % solved: 91.5% -> 94% | Helped |
| Use nine repeats without stacking instead of DRC(3,3) (ablation) | DRC architecture · Sokoban unfiltered | % solved | Hurt |
| Replace spatial ConvLSTM modules with flat 1D LSTMs (ablation) | DRC(3,3) · Sokoban | % solved | Hurt |
| Remove pool-and-inject (ablation) | DRC(3,3) · Sokoban | % solved | Mixed |
Closest papers by meaning
- Interpreting Emergent Planning in Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.91)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.90)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.89)
- Reinforcement learning with combinatorial actions for coupled restless bandits (ICLR 2025 · similarity 0.88)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.88)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · 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.