Research map / Vision-language and multimodal / Robot action and navigation
Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation
Xiaohuan Pei, Yuxing Chen, Siyu Xu, Yunke Wang, Yuheng Shi, Chang Xu
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 action-aware dynamic gating to text-driven pruning (ablation) | VLA-ADP on OpenVLA-OFT · LIBERO average | success rate: 93.45% -> 96.3% | Improved |
| use dynamic gating instead of periodic switching (ablation) | VLA-ADP on OpenVLA-OFT · LIBERO-Object | success rate: 81.4% -> 98.0% | Improved |
| use dynamic gating instead of periodic switching (ablation) | VLA-ADP on OpenVLA-OFT · LIBERO average | success rate: 89.9% -> 96.3% | Improved |
| score tokens at deeper layers (1 or 4) instead of layer 0 (ablation) | VLA-ADP on OpenVLA-OFT · LIBERO average | success rate: 96.3% -> 95.5% (layer 1), 95.8% (layer 4) | Hurt |
| score tokens at layer 4 instead of layer 0 (ablation) | VLA-ADP on OpenVLA-OFT · LIBERO average | FLOPs: 6.43 -> 6.89 | Hurt |
| text-driven relevance pruning instead of random token dropping (50%) | OpenVLA-OFT · LIBERO-Object | success rate: 73.0% -> 98.0% | Improved |
| apply ADP at keep ratio 50-70% to OpenVLA-OFT | OpenVLA-OFT · LIBERO average | success rate: 97.1% -> 96.3% with 1.23x speedup | Mixed |
| apply ADP instead of FastV pruning on OpenVLA-OFT | OpenVLA-OFT · LIBERO average | success rate: 86.8% -> 96.3% | Improved |
| apply ADP to OpenVLA-OFT on real robot | OpenVLA-OFT · Real-world (Jaco2, 4 tasks) | average success rate: 85.8% -> 88.3%, latency 76.9 -> 51.8 | Improved |
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 1 of 1 shared result
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
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.93)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.92)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.91)
- When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA (NeurIPS 2026 · similarity 0.90)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (NeurIPS 2026 · 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.
