Research map / Vision-language and multimodal / Robot action and navigation
Contrastive Representation Regularization for Vision-Language-Action Models
Taeyoung Kim, Jimin Lee, Myungkyu Koo, Dongyoung Kim, Kyungmin Lee, Changyeon Kim, Younggyo Seo, Jinwoo Shin
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 RS-CL regularization to GR00T N1.5 | GR00T N1.5 · RoboCasa-Kitchen (300 demos) | success rate (%): 65.7 -> 69.7 | Improved |
| Add RS-CL to GR00T N1.5 | GR00T N1.5 · LIBERO average | success rate (%): 95.7 -> 96.4 | Improved |
| Add RS-CL to GR00T N1.5 | GR00T N1.5 · LIBERO Object | success rate (%): 99.4 -> 98.6 | Hurt |
| Add RS-CL to GR00T N1.5 | GR00T N1.5 · RoboCasa-Kitchen pick-and-place tasks | success rate (%): 30.3% to 41.5% (+11.2%) | Improved |
| Add RS-CL to GR00T N1.5 | GR00T N1.5 · RoboCasa-Kitchen open-and-close (30 demos) | success rate (%): 71.3 -> 70.0 | No clear effect |
| Add RS-CL to GR00T N1.5 | GR00T N1.5 · Real-robot tasks | success rate (%): 45.0% to 58.3% (+13.3%) | Improved |
| Use RS-CL instead of single-view TCN auxiliary loss | GR00T N1.5 · RoboCasa-Kitchen (30 demos) | success rate (%): 50.3 -> 53.0 | Improved |
| Use RS-CL instead of TCN regarding compute | GR00T N1.5 · RoboCasa-Kitchen (30 demos) | FLOPs (x10^12): 7.53 -> 2.91 | Improved |
| Add contrastive regularization (InfoNCE) without soft labels (ablation) | GR00T N1.5 · RoboCasa-Kitchen (300 demos) | success rate (%): 65.7 -> 67.3 | Improved |
| Use current-state distance soft labels instead of InfoNCE (ablation) | RS-CL · RoboCasa-Kitchen (300 demos) | success rate (%): 67.3 -> 69.7 | Improved |
| Use next action sequence distance instead of state distance as soft-label target (ablation) | RS-CL · RoboCasa-Kitchen (300 demos) | success rate (%): 69.7 -> 66.7 | Hurt |
| Add view cutoff augmentation instead of no augmentation (ablation) | RS-CL · RoboCasa-Kitchen (300 demos) | success rate (%): 65.3 -> 69.7 | Improved |
Closest papers by meaning
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.90)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (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.
