3 papers indexed
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…
Alessandro Canevaro, Hang Yu, Julian Schmidt, Peizheng Li, Silvan Lindner, Wilhelm Stork, et al.
While closed-loop motion planners trained on large-scale, object-level datasets, e.g., nuPlan, demonstrate strong in-distribution (ID) performance, their generalization to novel urban topologies and recovery mechanisms following execution perturbations remain under-explored. To a…
Shijie Li, Yilin Gao, Siyuan Yang, Tieyuan Chen, Chaofan Gan, Zhihao He, et al.
Multimodal Large Language Models (MLLMs) are often constrained by a language-space bottleneck, forcing complex visual reasoning into discrete tokens which can lose perceptual nuance. A promising alternative is continuous latent reasoning, where the goal is to discover implicit re…