Research map / Vision-language and multimodal / Robot action and navigation
TRM-VLA: Temporal-Aware Chain-of-Thought Reasoning and Memorization for Vision-Language-Action Models
Xiang Li, Ya-Li Li, Yuan Wang, Shengjin Wang
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 keyframe-triggered reasoning (KTR) (ablation) | TRM-VLA (CogACT backbone) · SimplerEnv-Bridge | avg success rate: 0.54 -> 0.65 | Improved |
| add granularity-adaptable context memory (GCM) on top of KTR (ablation) | TRM-VLA · SimplerEnv-Bridge | avg success rate: 0.65 -> 0.73 | Improved |
| add stage-wise temporal structuring within KTR (ablation) | TRM-VLA · SimplerEnv-Bridge | avg success rate: 0.54 -> 0.60 | Improved |
| add keyframe-based annotation within KTR (with STS) (ablation) | TRM-VLA · SimplerEnv-Bridge | avg success rate: 0.60 -> 0.65 | Improved |
| add temporal reasoning integration (cross-attention retrieval) to GCM (ablation) | TRM-VLA · SimplerEnv-Bridge | avg success rate: 0.70 -> 0.73 | Improved |
| add dynamic reasoning context buffer to GCM (ablation) | TRM-VLA · SimplerEnv-Bridge | avg success rate: 0.69 -> 0.73 | Improved |
| add KTR on real-world tasks (ablation) | TRM-VLA · Real-world (4 tasks) | avg success rate: 0.50 -> 0.59 | Improved |
| add GCM on real-world tasks (ablation) | TRM-VLA · Real-world (4 tasks) | avg success rate: 0.59 -> 0.69 | Improved |
| add GCM on Push Buttons in Order and Place Bear (ablation) | TRM-VLA with KTR · Real-world Push Button | success rate: 0.45 -> 0.6 | Improved |
| trigger CoT only at keyframes instead of full CoT every step | TRM-VLA vs CogACT-ECoT · Real-world tasks | CoT tokens per step: 26.3 -> 4.3 | Improved |
| add full per-step CoT (ECoT) to CogACT-Base | CogACT-Base · Real-world tasks | avg success rate (%): 50.0 -> 44.0 | Hurt |
| add full per-step CoT (ECoT) to CogACT-Base | CogACT-Base · LIBERO-90 | success rate (%): 88.4 -> 92.1 | Improved |
Closest papers by meaning
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.92)
- Latent Reasoning VLA: Latent Thinking and Prediction for Vision-Language-Action Models (ICML 2026 · similarity 0.92)
- LaST$_{0}$: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Model (ICML 2026 · similarity 0.92)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.91)
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.90)
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.
