arxivcs.ROcs.AIcs.CV2026-07-10
TS-Mask VLA: 2D Temporal-Spatial Masking for Vision-Language-Action Model with Effective Bridging
Shengzhuo Yang, Ronghao Yu, Chuanjie Lv, Linpeng Peng, Hang Yu, Jie Ren, et al.
Vision-language-action (VLA) models aim to understand natural-language instructions and visual observations, and to generate and execute corresponding actions as embodied agents. Recently, autoregressive token-based action generation has driven the development of many representat…