Research map / Architectures and efficiency / Pruning and efficient tuning
EcoVLA: Environment-Aware Adaptive Pruning with Interleaved Inference Orchestration for Vision-Language-Action Models
Yuting Huang, Leilei Ding, Zhipeng Tang, Zenghuan Zhu, Jiajun Deng, Xinrui Lin, Shuo Liu, Haojie Ren, Jianmin Ji, Yanyong Zhang
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 static Wanda pruning with EcoVLA environment-aware adaptive pruning (25%) | OpenVLA-OFT · LIBERO-Goal | success rate (%): 87.2 -> 94.6 | Improved |
| replace static Wanda pruning with EcoVLA environment-aware adaptive pruning (40%) | OpenVLA-OFT · LIBERO-Goal | success rate (%): 77.0 -> 89.4 | Improved |
| replace Wanda with EcoVLA (40%) | OpenVLA-OFT · LIBERO average | success rate (%): 88.8 -> 94.0 | Improved |
| replace Wanda with EcoVLA (25%) | OpenVLA-OFT · LIBERO latency | latency (ms): 124.32 -> 113.98 | Improved |
| combine EcoVLA with FastV (25%) | FastV on OpenVLA-OFT · LIBERO average | success rate (%): 94.4 -> 96.2 | Improved |
| combine EcoVLA with FastV (25%) | FastV on OpenVLA-OFT · LIBERO | speedup: 1.21x -> 2.18x | Improved |
| combine EcoVLA with VLA-Cache (25%) | VLA-Cache on OpenVLA-OFT · LIBERO | speedup: 1.10x -> 1.34x | Improved |
| combine EcoVLA with VLA-Cache (25%) | VLA-Cache on OpenVLA-OFT · LIBERO average | success rate (%): 96.6 -> 95.5 | Hurt |
| increase pruning ratio from 25% to 40% | EcoVLA on OpenVLA-OFT · LIBERO average | success rate (%): 96.8 -> 94.0 | Hurt |
| increase pruning ratio from 25% to 37.5% | EcoVLA on pi0.5 · LIBERO-Long | success rate (%): 91.6 -> 87.0 | Hurt |
| apply EcoVLA on real robot | pi0.5 on Kinova Gen3 · Task2 success | successes out of 20: 18/20 -> 16/20 | Hurt |
| run pruning in parallel stream (I2O) instead of ignoring overhead | EcoVLA · OpenVLA-OFT latency | latency (ms): 143.56 -> 148.06 | Mixed |
Closest papers by meaning
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.92)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLA (NeurIPS 2026 · similarity 0.90)
- Each Complexity Deserves a Pruning Policy (NeurIPS 2025 · similarity 0.89)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (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.
