Research map / Vision-language and multimodal / Robot action and navigation
ReconVLA: Reconstructive Vision-Language-Action Model as Effective Robot Perceiver
Wenxuan Song, Ziyang Zhou, Han Zhao, Jiayi Chen, Pengxiang Ding, Haodong Yan, Yuxin Huang, Feilong Tang, Donglin Wang, Haoang Li
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 gaze-region reconstruction with pretraining (implicit grounding) to the VLA baseline | LLaVA-7b VLA · CALVIN ABC→D | Avg. Len: 3.36 -> 3.95 | Improved |
| add explicit grounding (cropped image as extra input) to the baseline | baseline VLA · CALVIN ABC→D | Avg. Len: 3.36 -> 3.61 | Improved |
| add chain-of-thought bounding box output before actions | baseline VLA · CALVIN ABC→D | Avg. Len: 3.36 -> 0.63 | Hurt |
| add large-scale visual reconstruction pretraining (ablation) | ReconVLA · CALVIN ABC→D | Avg. Len: 3.85 -> 3.95 | Improved |
| reconstruct only the gaze region instead of the entire image (ablation) | ReconVLA (no pretraining) · CALVIN ABC→D | Avg. Len: 3.42 -> 3.85 | Improved |
| add whole-image reconstruction auxiliary loss (ablation) | baseline VLA (no pretraining) · CALVIN ABC→D | Avg. Len: 3.36 -> 3.42 | Improved |
| add gaze-region reconstruction without pretraining (ablation) | baseline VLA · CALVIN ABC→D | success rate 5/5 (%): 49.0 -> 58.2 | Improved |
| add gazing mechanism on stack block task | baseline VLA · CALVIN ABC→D, stack block | success rate (%): 59.3% -> 79.5% | Improved |
| ReconVLA vs strongest generative/VLA baselines | ReconVLA · CALVIN ABCD→D | Avg. Len: 4.21 -> 4.23 | Improved |
| ReconVLA on unseen real-world objects vs OpenVLA and PD-VLA | ReconVLA · real-world unseen tasks | success rate: baselines nearly 0% | Improved |
Later papers that compared with this method
2 later accepted papers report ReconVLA as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Unified Diffusion VLA: Vision-Language-Action Model via Joint Discrete Diffusion Diffusion Process (ICLR 2026 · better on 1 of 1 shared result)
- OASIS: Observation-Action Space Alignment via SE(3) Trajectory Prediction for Robotic Manipulation (NeurIPS 2026 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VLAS — VLAS: Vision-Language-Action Model with Speech Instructions for Customized Robot Manipulation (ICLR 2025) · better on 1 of 1 shared result
- CLOVER — Clover: Towards a Unified Video-Language Alignment and Fusion Model (CVPR 2023) · 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
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.90)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (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.
