Research map / Vision-language and multimodal / Robot action and navigation
TTF-VLA: Temporal Token Fusion via Pixel-Attention Integration for Vision-Language-Action Models
Chenghao Liu, Jiachen Zhang, Chengxuan Li, Zhimu Zhou, Shixin Wu, Songfang Huang, Huiling Duan
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 Temporal Token Fusion to OpenVLA | OpenVLA · LIBERO average | success rate (%): 68.4 -> 72.4 | Improved |
| add TTF to VLA-Cache | VLA-Cache · LIBERO average | success rate (%): 71.3 -> 74.0 | Improved |
| add TTF to VLA-Cache on Spatial suite | VLA-Cache · LIBERO Spatial | success rate (%): 84.0 -> 84.0 | No clear effect |
| add TTF to base OpenVLA-7B without fine-tuning | OpenVLA · SimplerEnv average | success rate (%): 33.2 -> 34.9 | Improved |
| add TTF to fine-tuned OpenVLA in real robot tasks | OpenVLA · Real robot average | success rate (%): 38.3 -> 41.7 | Improved |
| add TTF on drawer-closing real robot task | OpenVLA · Real robot Drawer | success rate (%): 45.0 -> 45.0 | No clear effect |
| use pixel-difference detection only instead of dual detection (ablation) | TTF · LIBERO average | success rate (%): 72.4 -> 70.4 | Hurt |
| use attention detection only instead of dual detection (ablation) | TTF · LIBERO average | success rate (%): 72.4 -> 71.3 | Hurt |
| pixel-only TTF on Spatial and Goal suites (ablation) | TTF · LIBERO Spatial/Goal | success rate (%): Spatial 82.0 -> 80.5; Goal 77.0 -> 76.5 | Hurt |
| increase keyframe interval K beyond 30 (ablation) | TTF · LIBERO Object/Long | success rate (%) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VLA-Cache — VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.91)
- $\text{A}^3$-VLA: Automatic Perception-Guided Attention Alignment for Robot Manipulation (NeurIPS 2026 · similarity 0.91)
- SemanticVLA: Semantic-Aligned Sparsification and Enhancement for Efficient Robotic Manipulation (AAAI 2026 · similarity 0.91)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.89)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · 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.
