Research map / Label-efficient and robust learning / Domain adaptation and generalization
Domain Arithmetic: One-Shot VLA Adaptation under Environmental Shifts
Taewook Kang, Taeheon Kim, Donghyun Shin, Jonghyun Choi
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 domain vector to the base policy | pi0.5 · LIBERO novel viewpoints | success rate: 24.6 percentage points | Improved |
| DART instead of FLA (adapt whole model, not vision encoder) | pi0.5 · LIBERO novel viewpoints | average success rate: 74.3 -> 79.1 | Improved |
| DART instead of RETAIN | pi0.5 · LIBERO novel viewpoints | average success rate: 69.6 -> 79.1 | Improved |
| one-shot fine-tuning instead of zero-shot | pi0.5 · LIBERO novel viewpoints | average success rate: 54.5 -> 31.5 | Hurt |
| DART instead of best baseline on real robot | pi0.5 · Real-world UR10e novel viewpoint | average success rate: 55.0 -> 81.7 | Improved |
| DART instead of zero-shot on pi0-FAST | pi0-FAST · LIBERO novel viewpoints | average success rate: 73.4 -> 79.4 | Improved |
| DART instead of zero-shot cross-embodiment | pi0.5 · MimicGen Panda to UR5e (cross-embodiment) | average success rate: 62.0 -> 69.4 | Improved |
| add subspace alignment (filtering and scaling) (ablation) | DART · LIBERO novel viewpoints | success rate | Improved |
| DART without subspace alignment instead of one-shot fine-tuning (ablation) | DART w/o SA · LIBERO novel viewpoints | success rate | Improved |
| randomized SVD (r=256) instead of full SVD (ablation) | DART · LIBERO novel viewpoints | average success rate: 79.1 -> 78.7 | No clear effect |
| randomized SVD (r=256) instead of full SVD (ablation) | DART · LIBERO novel viewpoints | runtime: 15m 35s -> 6m 33s | Improved |
| increase scaling coefficient alpha too far (ablation) | DART · LIBERO novel viewpoints | success rate | Hurt |
Closest papers by meaning
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.86)
- OTTER: A Vision-Language-Action Model with Text-Aware Visual Feature Extraction (ICML 2025 · similarity 0.86)
- Prompt-Based Distribution Alignment for Unsupervised Domain Adaptation (AAAI 2024 · similarity 0.86)
- Enhancing Domain Adaptation through Prompt Gradient Alignment (NeurIPS 2024 · similarity 0.86)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (NeurIPS 2026 · similarity 0.86)
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.
