Research map / Vision-language and multimodal / Robot action and navigation
See What Matters: Differentiable Grid Sample Pruning for Generalizable Vision-Language-Action Model
Yixu Feng, Zinan Zhao, Yanxiang Ma, Chenghao Xia, Chengbin Du, Yunke Wang, 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 |
|---|---|---|---|
| replace post-training discrete pruning (FastV, SparseVLM) with jointly trained GridS16 | π0 · LIBERO average | success rate (%): 94.4 -> 96.0 | Improved |
| reduce visual tokens from 256 to 4 with GridS (ablation) | π0.5 · LIBERO average | success rate (%): 96.7 -> 96.7 | No clear effect |
| reduce visual tokens to 16 with GridS | π0 · LIBERO FLOPs | FLOPs (G): 216.01 -> 51.65 | Improved |
| reduce GridS token budget to K=4 (extreme sparsity) (ablation) | π0.5 · LIBERO-Spatial and Long | success rate (%): Long 95.2 -> 92.8 | Hurt |
| reduce GridS token budget to K=4 on Object suite (ablation) | π0 · LIBERO-Object | success rate (%): 99.2 -> 99.4 | Improved |
| use GridS with 16 tokens in place of 256 on ALOHA | π0 · ALOHA Insertion (Human) | accuracy (%): 56.7 -> 64.2 | Improved |
| use GridS with 16 tokens on ALOHA Insertion (Scripted) | π0 · ALOHA Insertion (Scripted) | accuracy (%): 91.4 -> 86.9 | Hurt |
| add GridS (16 tokens) to SmolVLA on real robot | SmolVLA · Real-world Stack Cubes | success rate (%): 9.5 -> 61.9 | Improved |
| vary sampled token count K from 4 to 8 to 16 to 32 (ablation) | GridS on SmolVLA · Real-world Stack Cubes | success rate (%): 0.0 / 28.5 / 61.9 / 19.0 | Mixed |
| add GridS (K=16) for OOD evaluation | SmolVLA · Real-world Stack Cubes OOD (21 scenarios) | OOD success rate (%): 0.0 -> 38.1 | Improved |
| add GridS to training | π0 · training time per step (batch 128) | time (s/step): 14.32 -> 4.21 | Improved |
| add GridS (K=16) under wrist-only inference with token masking | π0 on LIBERO · LIBERO wrist-only (Long) | accuracy: 0.30 -> 0.70 | Improved |
Closest papers by meaning
- When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA (NeurIPS 2026 · similarity 0.90)
- FASTer: Toward Powerful and Efficient Autoregressive Vision–Language–Action Models with Learnable Action Tokenizer and Block-wise Decoding (ICLR 2026 · similarity 0.90)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.89)
- Balanced Token Pruning: Accelerating Vision Language Models Beyond Local Optimization (NeurIPS 2025 · similarity 0.89)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.89)
- ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models (CVPR 2025 · similarity 0.88)
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.
