Research map / Reinforcement learning / Visual and representation RL
TraversRL: Traversable Pedestrian Pathway Generation With Reinforcement Learning
Bin Han, Robert Wolfe, Bill Howe
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 |
|---|---|---|---|
| add RL fine-tuning with terminal graph reward | TraversRL-Pre · all datasets | buffered IoU: 5.1% on WashingtonInter, 5.6% on Seattle, 5.2% on D.C. | Improved |
| add local stepwise reward shaping on top of global reward (ablation) | TraversRL-Global · all datasets | buffered IoU: 1.6%, 4.1%, 2.1% | Improved |
| add local reward shaping | TraversRL-Global · D.C. (ViT-S) | buffered IoU: 0.226 -> 0.221 | No clear effect |
| RL fine-tuning reduces graph density | TraversRL-Pre · Seattle | AvgDeg: 0.574 -> 0.519 | Improved |
| add RL fine-tuning (global reward) | TraversRL-Pre · WashingtonInter | TravSim: 0.595 -> 0.599 | Improved |
| add local shaping | TraversRL-Pre · Seattle | TravSim: 0.389 -> 0.411 | Improved |
| use larger evaluation buffer 4m vs 1m (ablation) | TraversRL-Global · WashingtonInter | IoU gain over Pre: 2.1% at 4m, 3.1% at 2m, 5.2% at 1m | Mixed |
| use stricter 1m buffer for local reward (ablation) | TraversRL-Local · WashingtonInter | IoU gain over Pre: 2.5% at 4m, 4.0% at 2m, 6.8% at 1m | Improved |
| increase action space from |A|=181 to 361 (ablation) | TraversRL-Local ResNet-18 · WashingtonInter | IoU/TravSim: 0.448/0.581 -> 0.540/0.605 | Improved |
| increase action space from |A|=361 to 721 (ablation) | TraversRL-Local ResNet-18 · WashingtonInter | IoU/TravSim: 0.540/0.605 -> 0.569/0.618 | Improved |
| increase action space from |A|=361 to 721 (ablation) | TraversRL-Local ResNet-18 · Seattle | IoU/TravSim: 0.199/0.403 -> 0.198/0.383 | Hurt |
| increase action space from |A|=361 to 721 (ablation) | TraversRL-Local ResNet-18 · D.C. | IoU/TravSim: 0.227/0.513 -> 0.237/0.520 | Improved |
Closest papers by meaning
- MapDreamer: Aerial Imagery Conditioned Latent Diffusion For Lane Level Map Generation (ECCV 2026 · similarity 0.85)
- WalkGPT: Grounded Vision-Language Conversation with Depth-Aware Segmentation for Pedestrian Navigation (CVPR 2026 · similarity 0.85)
- MapTracker: Tracking with Strided Memory Fusion for Consistent Vector HD Mapping (ECCV 2024 · similarity 0.85)
- Controllable Road Marking Generation (NeurIPS 2026 · similarity 0.83)
- VLMs Trace Without Tracking: Diagnosing Failures in Visual Path Following (NeurIPS 2026 · similarity 0.83)
- Hierarchical Semantic-Augmented Navigation: Optimal Transport and Graph-Driven Reasoning for Vision-Language Navigation (NeurIPS 2025 · similarity 0.83)
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.
