SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2606.07100 · ICML · 2026

LARA: Latent Action Representation Alignment for Vision-Language-Action Models

Siyuan Huang, Baoxiong Jia, Jiangyong Huang, Mengya Liu, Jingze Zhang

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 6 functions out of this paper's own repositories and ran 5 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
lmy1001/LARA canonical 5 of 6
FunctionStatusWhere it lives
swish Ran lmy1001/LARA/LARA_full/lara_full/model/action_head/action_encoder.py
pointer only (licence: NONE) · get_code("0f786c407fb1ee4c")
collate Ran lmy1001/LARA/LARA_full/lara_full/model/transforms.py
pointer only (licence: NONE) · get_code("7a91a74e5c4b02cd")
formalize_language Ran lmy1001/LARA/LARA_full/lara_full/model/transforms.py
pointer only (licence: NONE) · get_code("50ec0f583ab4f1cd")
squeeze_dict_values Ran lmy1001/LARA/LARA_full/lara_full/model/policy.py
pointer only (licence: NONE) · get_code("b4a7f7844eef1bba")
unsqueeze_dict_values Ran lmy1001/LARA/LARA_full/lara_full/model/policy.py
pointer only (licence: NONE) · get_code("3dffd60cbc50e250")
build_eagle_processor Not yet run lmy1001/LARA/LARA_full/lara_full/model/transforms.py
pointer only (licence: NONE) · get_code("5e7999ebef31c22e")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

ment. This enables reciprocal benefits where LAMs learn with action trajectories to avoid spurious visual changes, while VLAs are regularized by forward dynamics learned within LAMs to reduce hallucinations of functionally ineffective trajectories. We demonstrate LARA's versatility and effectiveness for pre-training, post-training enhancement of pre-trained VLA models, and LAM refinement, achieving an average of ∼10%, ∼5%, and ∼15% improvement over 3 simulation and 1 meticulously designed real-world robotic manipulation benchmarks. The code is publicly available at https://github.com/lmy1001/LARA.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2606.07100")
get_code_for_paper("2606.07100")
have("2606.07100")

Connect an agent — have() is free.