arXiv Machine Learning October 16, 2026 By Mozes Jacobs, Aiden Swann, Mathilde Papillon, Siddharth Boppana, Fenil R. Doshi, Usha Bhalla, Thomas Icard, Matt Feiszli, Thomas McGrath, Vasudev Shyam, Demba Ba, Leon Bergen, Nina Miolane, Owen Lewis, Jack Merullo, Ekdeep Singh Lubana, Atticus Geiger, Matthew Kowal, Thomas Fel

A Phenomenology of Neural Geometry in Vision-Language-Action Models

Read the original on arXiv Machine Learning →

The paper investigates the geometric structure of task-relevant variables inside vision‑language‑action (VLA) policies for robotic manipulation. By combining controlled experiments, natural rollouts, linear probes, and activation interventions, the authors identify seven key concepts and show that their representations are largely multi‑dimensional, shaped by constraints such as joint limits and task geometry. These geometric patterns are consistent across simulated and real‑world VLA models, and the authors illustrate how steering along these geometries can be used to guide policy behavior.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

arXiv Computer Vision

Oct 22

GeoMem-VLA: Internalizing 3D Scene Understanding into Temporal Memory for Vision-Language-Action Models

arXiv:2610.29866v1 Announce Type: new Abstract: Geometric structure is essential for robotic manipulation and understanding the 3D world. Most VLA models rely on 2D representations of current observa...

By Qinfan Zhang, Yifan Luo, Haoyuan Shi, Guanlin Liu, Han Dong, Haozhe Shan, Yong Dai, Xiangcong Ren, Xiaozhu Ju

rag robotics multimodal benchmarks

arXiv AI

Jun 24

G$^3$VLA: Geometric inductive bias for Vision-Language-Action Models

arXiv:2606. 24472v1 Announce Type: cross Abstract: Vision-language-action (VLA) models have made rapid progress in generalist robot manipulation by harnessing semantic knowledge from pretrained vision-language backbones, but their visual tokens remain grounded in 2D image coordinates rather than the calibrated geometry of the robot's cameras -- a mismatch especially pronounced in multi-camera setups, where views are coupled by known intrinsics and extrinsics yet processed as independent images.

By Yue Peng, Yongzhe Zhao, Artur Habuda, Khuyen Pham, Yanheng Zhu, Tran Nguyen Le, Fares Abu-Dakka, Li Guo

rag robotics multimodal safety

Hugging Face Trending Papers

Oct 21

GeoMem-VLA: Internalizing 3D Scene Understanding into Temporal Memory for Vision-Language-Action Models

Geometric structure is essential for robotic manipulation and understanding the 3D world. Most VLA models rely on 2D representations of current observations, while geometry-aware approaches often requ...

rag robotics multimodal benchmarks

arXiv Computer Vision

Sep 1

Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Survey

arXiv:2508.13073v3 Announce Type: replace-cross Abstract: Robotic manipulation, a key frontier in robotics and embodied AI, requires precise motor control and multimodal understanding, yet traditiona...

By Rui Shao, Wei Li, Lingsen Zhang, Renshan Zhang, Zhiyang Liu, Ran Chen, Liqiang Nie

llms agents reinforcement-learning robotics multimodal benchmarks

arXiv AI

Jun 9

GEAR-VLA: Learning Geometry-Aware Action Representations for Generalizable Robotic Manipulation

arXiv:2606. 08530v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models achieve strong benchmark performance but still struggle in real-world deployment with unseen objects, background shifts, and different robot embodiments.

By Yuan Zhang, Shiqi Zhang, Yedong Shen, Shuai Dong, Jiajun Deng, Xin Zhang, Yuxuan Gao, Jiajia Wu, Xin Nie, Zhiyuan Cheng, Jianmin Ji, Yanyong Zhang, Xingyi Zhang, Jia Pan

robotics multimodal benchmarks

arXiv AI

Sep 25

World Action Agent: Harnessing VLMs for Robot Manipulation via World Action Rehearsal

World Action Agent (WAA) is a multi‑agent framework that lets vision‑language models (VLMs) directly pilot robots by operating within a visual action workspace. The workspace provides automatically selected contact views, editable action rehearsals, and in‑view correction to refine decisions before low‑level execution. WAA learns procedural skills from expert videos and human teaching, and its interaction traces can train smaller VLMs, achieving state‑of‑the‑art success on LIBERO‑Pro and improving out‑of‑domain performance on robosuite and Qwen3.5‑9B.

By Yehang Zhang, Haojian Huang, Yifan Chang, Jianchong Su, Bohan Zhou, Yingjie Xu, Wosong Chen, Tianhao Zhou, Chenxu Wang, Tianyi Zhang, Yangkai Wei, Wenqian Li, Shiyuan Deng, Yinchuan Li, Ying-Cong Chen, Zexi Li

llms agents robotics fine-tuning multimodal benchmarks

One email a morning, machine-written

Related stories

GeoMem-VLA: Internalizing 3D Scene Understanding into Temporal Memory for Vision-Language-Action Models

G$^3$VLA: Geometric inductive bias for Vision-Language-Action Models

GeoMem-VLA: Internalizing 3D Scene Understanding into Temporal Memory for Vision-Language-Action Models

Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Survey

GEAR-VLA: Learning Geometry-Aware Action Representations for Generalizable Robotic Manipulation

World Action Agent: Harnessing VLMs for Robot Manipulation via World Action Rehearsal