Research map / Vision-language and multimodal / Robot action and navigation
From Perception to Planning: Evolving Ego-Centric Task-Oriented Spatiotemporal Reasoning via Curriculum Learning
Xiaoda Yang, Yuxiang Liu, Shenzhou Gao, Can Wang, Jingyang Xue, Lixin Yang, Yao Mu, Tao Jin, Zhimeng Zhang, Shuicheng YAN, Zhou Zhao
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 stage 3 LongTag data after CoT and Tag stages (ablation) | EgoTSR (Qwen2.5-VL-7B) · Level 2 Long | accuracy (%): 55.8 -> 92.4 | Improved |
| train stages sequentially (S1+S2+S3) instead of mixing all data (ablation) | EgoTSR · Level 2 Long | average accuracy (%): 69.6 -> 92.4 | Improved |
| add Subtask Planner (ablation) | EgoTSR · Level 2 Long | accuracy (%): 81.1 -> 92.4 | Improved |
| add Subtask Planner (ablation) | EgoTSR · Level 2 Long | bidirectional gap: 4.3 -> 0.1 | Improved |
| add Tag stage after CoT stage (S1 to S1+S2) (ablation) | EgoTSR · Level 2 Long | bidirectional gap: 89.5 -> 3.1 | Improved |
| add Tag stage after CoT stage (S1 to S1+S2) (ablation) | EgoTSR · Level 2 Long | average accuracy (%): 46.9 -> 55.8 | Improved |
| Tag-only training (EgoTSR-Tag) without LongTag (ablation) | EgoTSR-Tag · Level 2 Long | average accuracy (%): 55.8 (below 60%) | Hurt |
| add Tag stage after CoT stage (ablation) | EgoTSR · Level 1 Short | average accuracy (%): 73.6 -> 86.8 | Improved |
| add LongTag stage after Tag stage (ablation) | EgoTSR · Level 1 Short | average accuracy (%): 86.8 -> 87.5 | No clear effect |
| CoT-only training (S1) without Tag (ablation) | EgoTSR-CoT · Level 2 Long | bidirectional gap: Gap 89.5, forward 91.7 vs inverse 2.2 | Hurt |
Closest papers by meaning
- EgoThinker: Unveiling Egocentric Reasoning with Spatio-Temporal CoT (NeurIPS 2025 · similarity 0.90)
- EvoEmpirBench: Dynamic Spatial Reasoning with Agent-ExpVer (AAAI 2026 · similarity 0.89)
- SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.89)
- OST-Bench: Evaluating the Capabilities of MLLMs in Online Spatio-temporal Scene Understanding (NeurIPS 2025 · similarity 0.89)
- Linear Mechanisms for Spatiotemporal Reasoning in Vision Language Models (ICLR 2026 · similarity 0.89)
- EgoMind: Activating Spatial Cognition through Linguistic Reasoning in MLLMs (CVPR 2026 · 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.
