Research map / Reinforcement learning / Visual and representation RL
Value Iteration Networks
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 |
|---|---|---|---|
| add value iteration planning module (VIN) vs reactive CNN policy | grid-world policy · Grid-world 28x28 | success rate: 97% vs 74.2% | Helped |
| add value iteration planning module (VIN) vs reactive CNN policy | grid-world policy · Grid-world 16x16 | success rate: 99.3% vs 87.6% | Helped |
| untie weights of the K recurrent VI layers (ablation) | VIN · Grid-world 16x16 (weight sharing) | success rate: 95.6% vs 99.3% | Hurt |
| add VIN planning module with TRPO and curriculum training | grid-world RL policy · Grid-world RL 16x16 | success rate: 82.5% vs 33.1% | Helped |
| add VIN planning module with TRPO and curriculum training | grid-world RL policy · Grid-world RL 8x8 | success rate: 90.9% vs 86.9% | Helped |
| add VI planning on a coarse category subgraph to reactive WebNav policy | WebNav reactive baseline · WebNav random start | successful runs out of 4000: 346 vs 304 | Helped |
| add VI planning on a coarse category subgraph to reactive WebNav policy | WebNav reactive baseline · WebNav root start | successful runs out of 2000: 1030 vs 1025 | No clear effect |
| add VI planning on Wikipedia for Schools graph to reactive WebNav policy | full Wikipedia WebNav · full Wikipedia top-1 / top-4 | top-1 and top-4 test error: top-1 error 50.562% vs 52.019%; top-4 error 26.055% vs 24.424% | Mixed |
| learn VI from raw terrain image instead of obstacle labels | Mars rover VIN · Mars navigation | success rate: 84.8% vs 90.3% best achievable with labeled obstacles | No clear effect |
Closest papers by meaning
- Scaling Value Iteration Networks to 5000 Layers for Extreme Long-Term Planning (ICML 2025 · similarity 0.91)
- Highway Value Iteration Networks (ICML 2024 · similarity 0.91)
- Gated Path Planning Networks (ICML 2018 · similarity 0.90)
- Information-based Value Iteration Networks for Decision Making Under Uncertainty (ICLR 2026 · similarity 0.90)
- Neural Algorithmic Reasoners are Implicit Planners (NeurIPS 2021 · similarity 0.89)
- Value Propagation Networks (ICLR 2019 · 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.