Research map / Video / Video-language understanding
Target-Bench: Can Video World Models Achieve Mapless Path Planning with Semantic Targets?
Dingrui Wang, Zhihao Liang, Hongyuan Ye, Zhexiao Sun, Zhaowei Lu, Yuchen Zhang, Yuyu Zhao, Yuan Gao, Marvin Seegert, Finn Rasmus Schäfer, Haotong Qin, Wei Li and 4 more
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 |
|---|---|---|---|
| fine-tune Wan2.2-TI2V-5B with LoRA on 325 robot scenarios | Wan2.2-5B · Target-Bench (VGGT) | WO: 0.084 -> 0.330 | Improved |
| add shifting-frame data augmentation to fine-tuning (ablation) | Wan2.2-5B-FT · Target-Bench (VGGT) | WO: 0.330 -> 0.394 | Improved |
| add shifting-frame data augmentation to fine-tuning (ablation) | Wan2.2-5B-FT · Target-Bench (SpaTracker) | WO: 0.300 -> 0.385 | Improved |
| add shifting-frame data augmentation to fine-tuning (ablation) | Wan2.2-5B-FT · Target-Bench (ViPE) | WO: 0.208 -> 0.214 | Improved |
| fine-tune Wan2.2-5B on small robot dataset to surpass best off-the-shelf model | Wan2.2-5B-FT-DA · Target-Bench (VGGT) | WO: 0.341 -> 0.394 | Improved |
| use implicit instead of explicit semantic targets | video world models · Target-Bench (VGGT) | WO: 0.349 -> 0.331 | No clear effect |
| shorten planning horizon from 8s to 6s (ablation) | Wan2.2-I2V-Plus · Target-Bench (VGGT) | WO: 0.290 -> 0.338 | Improved |
| shorten planning horizon from 6s to 4s (ablation) | Wan2.2-I2V-Flash · Target-Bench (VGGT) | WO: 0.337 -> 0.363 | Improved |
| shorten planning horizon from 8s to 4s (ablation) | Wan2.2-I2V-Flash · Target-Bench (VGGT) | MR: 38.75 -> 23.18 | Improved |
| use ViPE instead of VGGT as world-decoder (ablation) | gt_video · Target-Bench | WO: 0.862 -> 0.291 | Hurt |
| use SpaTracker instead of VGGT as world-decoder (ablation) | gt_video · Target-Bench | WO: 0.862 -> 0.623 | Hurt |
| use ViPE instead of VGGT as world-decoder (ablation) | Wan2.2-Plus · Target-Bench | WO: 0.290 -> 0.265 | Mixed |
Closest papers by meaning
- EA-WM: Event-Aware Generative World Model with Structured Kinematic-to-Visual Action Fields (NeurIPS 2026 · similarity 0.87)
- Video Language Planning (ICLR 2024 · similarity 0.87)
- WorldModelBench: Judging Video Generation Models As World Models (NeurIPS 2025 · similarity 0.86)
- RoboDreamer: Learning Compositional World Models for Robot Imagination (ICML 2024 · similarity 0.86)
- From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction (NeurIPS 2025 · similarity 0.86)
- DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge (NeurIPS 2025 · 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.
