Research map / Vision-language and multimodal / Robot action and navigation
TIC-VLA: A Think-in-Control Vision-Language-Action Model for Robot Navigation in Dynamic Environments
Zhiyu Huang, Yun Zhang, Johnson Liu, Rui Song, Chen Tang, Jiaqi Ma
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 |
|---|---|---|---|
| use KV-cache features instead of waypoint guidance (with latency modeling) (ablation) | TIC-VLA · DynaNav | success rate (%): 22.35 -> 47.06 | Improved |
| add latency-aware modeling and training (KV cache interface) (ablation) | TIC-VLA · DynaNav | success rate (%): 30.59 -> 47.06 | Improved |
| add latency-aware modeling and training (waypoint interface) (ablation) | TIC-VLA · DynaNav | success rate (%): 16.47 -> 22.35 | Improved |
| add RL fine-tuning with injected delays (ablation) | TIC-VLA · DynaNav | success rate (%): 47.06 -> 55.29 | Improved |
| add RL fine-tuning (ablation) | TIC-VLA, Unitree Go2 on RTX 4060 · real-world navigation | success rate: 0.70 -> 0.85 | Improved |
| run VLM asynchronously instead of blocking control (ablation) | TIC-VLA · DynaNav | success rate (%): 32.94 -> 55.29 | Improved |
| add explicit VLM reasoning at test time (ablation) | TIC-VLA (RL) · DynaNav | success rate (%): 40.00 -> 55.29 | Improved |
| add explicit VLM reasoning at test time (ablation) | TIC-VLA (RL) · DynaNav | collision rate (%): 25.88 -> 28.24 | Hurt |
| use 3s action horizon instead of 1s (ablation) | TIC-VLA (no RL) · DynaNav | success rate (%): 42.35 -> 47.06 | Mixed |
| use 3s action horizon instead of 5s (ablation) | TIC-VLA (no RL) · DynaNav | success rate (%): 40.00 -> 47.06 | Improved |
| add ego-motion offset to the action policy (ablation) | TIC-VLA (no RL) · DynaNav | success rate (%): 41.18 -> 47.06 | Improved |
| run TIC-VLA on A6000 instead of local RTX 4060 | TIC-VLA · real-world navigation | success rate: 0.85 -> 0.80 | No clear effect |
Closest papers by meaning
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better (NeurIPS 2025 · similarity 0.90)
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.90)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.90)
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.89)
- World–Value–Action Model: Implicit Planning for Vision–Language–Action Systems (NeurIPS 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.
