Research map / Vision-language and multimodal / Robot action and navigation
GAE: Unleashing Physical Potential of VLM with Generalizable Action Expert
Mingyu Liu, Zheng Huang, Xiaoyi Lin, Muzhi Zhu, Canyu Zhao, Yating Wang, Haoyi Zhu, Hao Chen, Chunhua Shen
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 GAE action expert instead of direct IK execution (ablation) | VLM waypoint predictor · RoboTwin (1k SFT steps) | success rate: 0.32 -> 0.56 | Improved |
| add GAE under input depth noise (50 cm) (ablation) | VLM waypoint predictor · RoboTwin tasks | success rate: 0.04 -> 0.21 | Improved |
| train GAE with action pre-training then point cloud fine-tuning (APPF) (ablation) | GAE · RoboTwin long horizon | success rate: 0.50 -> 0.60 | Improved |
| inject guidance-pose noise of scale 0.1 when training GAE (ablation) | GAE · RoboTwin long horizon | success rate: 0.43 -> 0.60 | Improved |
| increase guidance-pose noise scale from 0.1 to 0.5 (ablation) | GAE · RoboTwin long horizon | success rate: 0.60 -> 0.41 | Hurt |
| use B-spline instead of linear interpolation (ablation) | GAE · RoboTwin short horizon | success rate: 0.75 -> 0.82 | Improved |
| use kinetic keyframes instead of uniform keyframe sampling (ablation) | GAE · RoboTwin | success rate: 0.81 -> 0.82 (short), 0.61 -> 0.60 (long) | No clear effect |
| single-benchmark VLM finetuning instead of multi-benchmark (ablation) | GAE · RoboTwin | success rate: 0.82 -> 0.85 (short), 0.60 -> 0.58 (long) | Mixed |
| increase SFT steps from 0 to 4k (ablation) | Qwen2.5-VL · MMLU | accuracy: 70.1 -> 29.4 | Hurt |
| increase SFT steps beyond 1k when GAE is used (ablation) | VLM + GAE · RoboTwin | success rate: 0.56 -> 0.58 | No clear effect |
| use sparse 3D waypoint interface with GAE instead of baselines | GAE · SIMPLER Google Robot | overall success rate: 69.8% -> 74.1% | Improved |
| same, on Variant Aggregation Drawer | GAE · SIMPLER Google Robot Drawer (Variant Aggregation) | success rate: 29.7% -> 28.3% | No clear effect |
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.
- SoFAR — SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation (NeurIPS 2025) · better on 3 of 3 shared results
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
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- MolmoMotion: Forecasting Point Trajectories in 3D with Language Instruction (NeurIPS 2026 · similarity 0.87)
- Event-Grounded Sparse Autoencoders for Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.87)
- ActQuant: Sub-4-bit Action-Guided Quantization for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
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.
