Research map / Vision-language and multimodal / Robot action and navigation
Predicting Consequences and Reinforcing Navigation Policies with Latent World Models
Zengmao Wang, Wei Gao, Shuhan Shen
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 counterfactual action sequences (M=16 vs M=1) (ablation) | LWM-B-P · SCAND | ACC3: 34.19 -> 70.46 | Improved |
| add counterfactual action sequences (M=16 vs M=1) (ablation) | LWM-B-P · SCAND | position error (PE): 2.239 -> 0.890 | Improved |
| increase counterfactual sequences from M=1 to M=2 (ablation) | LWM-B-P · SCAND | ACC5: 23.83 -> 44.55 | Improved |
| increase counterfactual sequences from M=8 to M=16 (ablation) | LWM-B-P · SCAND | ACC5: 46.11 -> 58.54 | Improved |
| add counterfactual action sequences (M=2 vs M=1) (ablation) | LWM-S · LWM dataset | ACC3: 43.96 -> 89.36 | Improved |
| increase counterfactual sequences beyond M=4 (ablation) | LWM-S · LWM dataset | ACC3: 88.46 -> 87.20 | No clear effect |
| add counterfactual action sequences (M=2 vs M=1) (ablation) | LWM-B-P · LWM dataset | position error (PE): 5.205 -> 3.586 | Improved |
| increase counterfactual sequences beyond M=4 (ablation) | LWM-B-P · LWM dataset | ACC3: 62.88 -> 61.26 | No clear effect |
| use MSE with log label instead of contrastive loss (ablation) | LWM-B-P · LWM dataset | ACC3: 36.00 -> 90.60 | Improved |
| use MSE with log label instead of plain MSE (ablation) | LWM-B-P · LWM dataset | ACC3: 86.40 -> 90.60 | Improved |
| use MSE with log label instead of contrastive and ranking loss (ablation) | LWM-B-P · LWM dataset | ACC3: 85.40 -> 90.60 | Improved |
| provide goal prompt in advance (LWM-B-P vs LWM-B) (ablation) | LWM-B · SCAND | ACC3: 58.03 -> 83.93 | Improved |
Closest papers by meaning
- World Action Verifier: Self-Improving World Models via Forward-Inverse Asymmetry (NeurIPS 2026 · similarity 0.89)
- Simplifying Latent Dynamics with Softly State-Invariant World Models (NeurIPS 2024 · similarity 0.88)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.88)
- Navigation World Models (CVPR 2025 · similarity 0.87)
- Predictive but Not Plannable: RC-aux for Latent World Models (NeurIPS 2026 · similarity 0.87)
- Learning World Models for Unconstrained Goal Navigation (NeurIPS 2024 · similarity 0.87)
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.
