Research map / Vision-language and multimodal / Robot action and navigation
SemanticVLA: Semantic-Aligned Sparsification and Enhancement for Efficient Robotic Manipulation
Wei Li, Renshan Zhang, Rui Shao, Zhijian Fang, Kaiwen Zhou, Zhuotao Tian, Liqiang Nie
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 ID-Pruner on SigLIP and SA-Pruner on DINOv2 instead of ID-Pruner on both (ablation) | SD-Pruner · LIBERO | overall success rate (%): 91.9 -> 97.1 | Improved |
| use ID-Pruner on SigLIP and SA-Pruner on DINOv2 instead of SA-Pruner on both (ablation) | SD-Pruner · LIBERO | overall success rate (%): 94.6 -> 97.1 | Improved |
| assign pruners correctly rather than swapped (SA on SigLIP, ID on DINOv2) (ablation) | SD-Pruner · LIBERO | overall success rate (%): 95.0 -> 97.1 | Improved |
| add HF-Fuser (ablation) | SemanticVLA without SA-Coupler · LIBERO | overall success rate (%): 93.6 -> 95.6 | Improved |
| add SA-Coupler (ablation) | SemanticVLA without HF-Fuser · LIBERO | overall success rate (%): 93.6 -> 94.1 | Improved |
| add both HF-Fuser and SA-Coupler (ablation) | baseline VLA with pruning · LIBERO | overall success rate (%): 93.6 -> 97.1 | Improved |
| raise sparsification ratio from 8x to 16x (ablation) | SemanticVLA · LIBERO | overall success rate (%): 97.7 -> 95.8 | Hurt |
| raise sparsification ratio from 8x to 32x (ablation) | SemanticVLA · LIBERO | overall success rate (%): 97.7 -> 92.0 | Hurt |
| lower sparsification ratio from 8x to 4x (ablation) | SemanticVLA · LIBERO | FLOPs (T): 2.37 T -> 3.28 T with SR unchanged at 97.7 | Hurt |
| use instruction-aware pruning instead of FastV at 8x compression (ablation) | visual token pruning · LIBERO | overall success rate (%): 88.8 -> 97.7 | Improved |
| use instruction-aware pruning instead of SliME at 8x compression (ablation) | visual token pruning · LIBERO | overall success rate (%): 85.6 -> 97.7 | Improved |
| SemanticVLA vs OpenVLA-OFT on real robot | SemanticVLA · Real-world AgileX | overall success rate (%): 55.6% -> 77.8% | Improved |
Closest papers by meaning
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.91)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.91)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.90)
- ForceVLA: Enhancing VLA Models with a Force-aware MoE for Contact-rich Manipulation (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.
