Research map / Vision-language and multimodal / Robot action and navigation
SwiftVLA: Unlocking Spatiotemporal Dynamics for Lightweight VLA Models at Minimal Overhead
Chaojun Ni, Cheng Chen, Xiaofeng Wang, Zheng Zhu, Wenzhao Zheng, Boyuan Wang, Tianrun Chen, Guosheng Zhao, Haoyun Li, Zhehao Dong, Qiang Zhang, Yun Ye and 3 more
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 4D features as VLM input (ablation) | SwiftVLA · RoboTwin 2.0 | average SR: 0.36 -> 0.40 | Improved |
| add Fusion Tokens with future trajectory supervision (ablation) | SwiftVLA with 4D input · RoboTwin 2.0 | average SR: 0.40 -> 0.50 | Improved |
| add 4D feature masking during training (ablation) | SwiftVLA (2D-only inference) · RoboTwin 2.0 | average SR: 0.02 -> 0.40 | Improved |
| add feature reconstruction to 4D masking (ablation) | SwiftVLA (2D-only inference) · RoboTwin 2.0 | average SR: 0.40 -> 0.50 | Improved |
| add 2D feature masking (ablation) | SwiftVLA (2D-only inference) · RoboTwin 2.0 | average SR: 0.50 -> 0.53 | Improved |
| randomize temporal cache size K in {3,4,5,6} during training instead of fixed (ablation) | SwiftVLA · RoboTwin 2.0 | average SR: 0.52 -> 0.53 | Improved |
| increase fixed cache size from K=3 to K=6 (ablation) | SwiftVLA · RoboTwin 2.0 | average SR: 0.47 -> 0.52 | Improved |
| use 4D input at inference in addition to mask-and-reconstruct training | SwiftVLA · RoboTwin 2.0 | average SR: 0.53 -> 0.55 | No clear effect |
| use SwiftVLA instead of SmolVLA | lightweight VLA · RoboTwin 2.0 | average SR: 0.29 -> 0.53 | Improved |
| use SwiftVLA instead of SmolVLA | lightweight VLA · real-world tasks | average score: 0.34 -> 0.80 | Improved |
| use SwiftVLA instead of pi0 on edge device | VLA · Jetson Orin | inference time (s): 2.966 -> 0.167 | Improved |
| use SwiftVLA instead of OpenVLA-OFT | VLA · LIBERO average | success rate: 97.1 -> 94.7 | Hurt |
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.
- 4D-VLA — 4D-VLA: Spatiotemporal Vision-Language-Action Pretraining with Cross-Scene Calibration (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
- 4D-VLA: Spatiotemporal Vision-Language-Action Pretraining with Cross-Scene Calibration (NeurIPS 2025 · similarity 0.90)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.89)
- 3D-VLA: A 3D Vision-Language-Action Generative World Model (ICML 2024 · similarity 0.89)
- ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation (CVPR 2026 · similarity 0.89)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.88)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · 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.
