Event
X-World: Transferring Video Priors to Variable-View Robot Control
1 reports · 1 sources
Chun Gu, Xiaofei Wei, Yiming Chen, Juntong Zhang, and Li Zhang released the X-World framework, which transfers video priors to a shared action expert via fixed-length token sequences and extends a pretrained video backbone to one, two, or three views using residual cross-view attention while preserving the original self-attention path. Training first separates video adaptation from pure action pretraining (the latter freezes the video backbone), then performs joint post-training in the target domain. The authors report: an average success rate of 98.5% on LIBERO; 76.5% on LIBERO-Plus with the LIBERO-trained checkpoint and no additional adaptation; and an average success rate of 81.7% across six dual-arm real-world tasks with 20 trials each. These numbers are all results self-reported by the paper's authors.
Report timeline
- X-World: Transferring video priors to variable-view robot control
X-World proposes a framework that transfers video priors to a shared action expert through fixed-length token sequences, and uses residual cross-view attention to extend a pretrained video backbone to one, two, or three views. The policy reaches 98.5% mean success on LIBERO, and its LIBERO-trained checkpoint obtains 76.5% on LIBERO-Plus without additional adaptation; across six dual-arm real-world tasks, it achieves a mean success rate of 81.7% over 20 trials per task.
arXiv cs.RO ·