Research map / Vision-language and multimodal / Robot action and navigation
Hybrid Training for Vision-Language-Action Models
Pietro Mazzaglia, Cansu Sancaktar, Markus Peschl, Daniel Dijkman
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 |
|---|---|---|---|
| train with Hybrid Training (HyT) instead of standard OpenVLA fine-tuning | OpenVLA 7B · Real-world tasks overall | success rate: 41% -> 63% | Improved |
| train with HyT instead of OpenVLA | OpenVLA 7B · Real-world out-of-distribution tasks | success rate: 29% -> 54% | Improved |
| train with HyT instead of OpenVLA | OpenVLA 7B · Real-world in-distribution tasks | success rate: 52% -> 72% | Improved |
| train with HyT on top of VLA-OFT, Prismatic initialization | VLA-OFT · LIBERO Goal and Long | success rate | Improved |
| train with HyT instead of VLA-OFT starting from OpenVLA | VLA-OFT · LIBERO (OpenVLA init) | overall success: 95.3% | No clear effect |
| train with HyT instead of standard VLA training at all data scales | VLA · ClevrSkills | success rate | Improved |
| use ECoT-style chain-of-thought generation at inference instead of HyT act mode | VLA · ClevrSkills | inference speed: 3x slower | Hurt |
| use HiRobot hierarchical generation instead of HyT act mode | VLA · ClevrSkills | inference speed: 4x slower | Hurt |
| replace self-predicted thoughts with oracle thoughts at inference | HyT · ClevrSkills complex tasks | success rate | Improved |
| add the think distribution to training coefficients (ablation) | HyT · ClevrSkills (300 demos) | success rate | Improved |
| add the follow distribution to training coefficients (ablation) | HyT · ClevrSkills oracle instruction following | success rate | Improved |
| add the follow distribution to training coefficients (acting mode alone) (ablation) | HyT · ClevrSkills acting mode | success rate | No clear effect |
Closest papers by meaning
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.91)
- HALO: A Unified Vision-Language-Action Model for Embodied Multimodal Chain-of-Thought Reasoning (ICML 2026 · similarity 0.90)
- Unifying Perception and Action: A Hybrid-Modality Pipeline with Implicit Visual Chain-of-Thought for Robotic Action Generation (CVPR 2026 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge (NeurIPS 2025 · 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.
