Newest papers

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

Qinfan Zhang, Yifan Luo, Haoyuan Shi, Guanlin Liu, Han Dong, Haozhe Shan, Yong Dai, Xiangcong Ren · 2026 · arXiv

DOI: 10.48550/arXiv.2610.29866 · open-access PDF

Indonesian

Struktur geometris sangat penting untuk manipulasi robotik dan memahami dunia 3D. Kebanyakan model VLA bergantung pada representasi 2D dari observasi saat ini, sementara pendekatan geometri sadar sering membutuhkan informasi 3D tambahan. Namun, geometri yang diduga independen dari observasi individu sering dikeluarkan daripada terkumpul menjadi representasi tugas yang persisten. wawasan utama kami adalah bahwa generasi aksi membutuhkan abstraksi geometris yang kompak dan relevan dengan tugas yang dapat diperbarui secara terus-menerus, bukan rekonstruksi 3D yang lengkap dan dapat ditampilkan pada setiap langkah.

Machine translation (Meta M2M100); the English original follows.

Abstract

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 require additional 3D information. However, geometry inferred independently from individual observations is often discarded rather than accumulated into a persistent task representation. Our key insight is that action generation needs a compact, task-relevant geometric abstraction that can be continuously updated, rather than a complete, renderable 3D reconstruction at every step. We introduce GeoMem-

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange