Research map / Reinforcement learning / Visual and representation RL
No RL, No Simulation: Learning to Navigate without Navigating
Meera Hahn, Devendra Singh Chaplot, Shubham Tulsiani, Mustafa Mukadam, James Matthew Rehg, Abhinav Gupta
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 |
|---|---|---|---|
| replace predicted distance network G_D with ground truth (ablation) | NRNS · Gibson Straight Hard | Success @1m: 23.45 -> 62.16 | Improved |
| replace predicted distance network G_D with ground truth (ablation) | NRNS · Gibson Curved Hard | Success @1m: 12.50 -> 47.70 | Improved |
| replace predicted target network G_T with ground truth (ablation) | NRNS · Gibson Straight Hard | Success @1m: 62.16 -> 95.91 | Improved |
| replace predicted exploration module G_EA with ground truth (ablation) | NRNS · Gibson Straight Hard | Success @1m: 95.91 -> 100.00 | Improved |
| replace G_D with ground truth (ablation) | NRNS · MP3D Straight Hard | Success @1m: 22.30 -> 58.40 | Improved |
| inject sensor and actuation noise | NRNS · Gibson Straight Easy | Success @1m: 68.00 -> 64.10 | Hurt |
| inject sensor and actuation noise | NRNS · Gibson Straight Hard | Success @1m: 23.82 -> 25.19 | Improved |
| use GRU memory instead of metric map in BC | BC baseline · Gibson Straight Easy | Success @1m: 24.80 -> 34.90 | Improved |
| train RL with 5x data and 100M steps instead of 10M steps | RL baseline · Gibson Straight Easy | Success @1m: 10.50 -> 43.20 | Improved |
| NRNS vs best BC baseline | NRNS · Gibson Straight Easy | Success @1m: 34.90 -> 68.00 | Improved |
| train NRNS on RealEstate10K instead of Gibson videos | NRNS · Gibson Straight Easy | Success @1m: 68.00 -> 56.42 | Hurt |
| train NRNS on RealEstate10K instead of Gibson videos | NRNS · Gibson Curved Medium | Success @1m: 23.90 -> 12.90 | Hurt |
Closest papers by meaning
- NavOL: Navigation Policy with Online Imitation Learning (ICML 2026 · similarity 0.88)
- Learning to Move with Affordance Maps (ICLR 2020 · similarity 0.88)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · similarity 0.87)
- Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL (NeurIPS 2024 · similarity 0.87)
- Stabilizing Contrastive RL: Techniques for Robotic Goal Reaching from Offline Data (ICLR 2024 · similarity 0.86)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.86)
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.
