Research map / Vision-language and multimodal / Robot action and navigation
LARA: Latent Action Representation Alignment for Vision-Language-Action Models
Mengya Liu, Baoxiong Jia, Jiangyong Huang, Jingze Zhang, Siyuan Huang
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 LARA latent action representation alignment to a vanilla DiT VLA (ablation) | LARA (DiT-only) · LIBERO (OXE-constrained) average | success rate (%): 84.4 -> 88.6 | Improved |
| add LARA alignment to a vanilla DiT VLA (ablation) | LARA (DiT-only) · SIMPLER-ENV (OXE-constrained) average | success rate (%): 55.8 -> 65.2 | Improved |
| add LARA alignment to a vanilla DiT VLA (ablation) | LARA (DiT-only) · LIBERO-Long (OXE-constrained) | success rate (%): 76.5 -> 86.0 | Improved |
| add LARA alignment to a vanilla DiT VLA (ablation) | LARA (DiT-only) · SIMPLER-ENV Drawer (OXE-constrained) | success rate (%): 21.0 -> 29.5 | Improved |
| add LARA alignment to a vanilla DiT VLA (ablation) | LARA (DiT-only) · SIMPLER-ENV Move (OXE-constrained) | success rate (%): 84.0 -> 83.7 | No clear effect |
| post-train GR00T-N1.6 with LARA using an OXE-pretrained LAM | GR00T-N1.6 · LIBERO (unconstrained) average | success rate (%): 95.0 -> 95.6 | Improved |
| post-train GR00T-N1.6 with LARA | GR00T-N1.6 · SIMPLER-ENV (unconstrained) average | success rate (%): 78.9 -> 79.9 | Improved |
| replace vanilla LAM with LARA-refined LAM as pseudo-labels | Moto-GPT · SIMPLER-ENV (Moto-GPT, OXE Fractal) average | success rate (%): 42.8 -> 49.5 | Improved |
| apply alignment at layer L instead of L-2 (ablation) | pi0.5-LARA · LIBERO average (pi0.5 post-training) | success rate (%): 91.2 -> 97.8 | Improved |
| apply LARA at final layer to pi0.5 | pi0.5 · LIBERO average (pi0.5 post-training) | success rate (%): 96.9 -> 97.8 | Improved |
| vary alignment depth over layers 4, 8, L-2, L (ablation) | GR00T-N1.6-LARA · LIBERO-Long (unconstrained) | success rate (%) | Mixed |
| jointly optimize LAM with the VLA instead of freezing the LAM (ablation) | GR00T-N1.6-LARA · LIBERO-Long (unconstrained) | success rate (%) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- villa-X — villa-X: Enhancing Latent Action Modeling in Vision-Language-Action Models (ICLR 2026) · better on 2 of 2 shared results
- CoT-VLA — CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025) · better on 1 of 1 shared result
- DreamVLA — DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledge (NeurIPS 2025) · 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
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.94)
- Latent Action Pretraining from Videos (ICLR 2025 · similarity 0.93)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- ATI-VLA: Action-Centric Predictive Vision–Language–Action Models via Actionable Alignment Then Adaptive Injection (NeurIPS 2026 · similarity 0.92)
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.92)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
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.
