6 papers indexed
Junsong Chen, Jincheng Yu, Yitong Li, Shuchen Xue, Haozhe Liu, Jingyu Xin, et al.
We introduce SANA-Video 2.0, a hybrid video diffusion transformer instantiated at 5B and 14B scales under a unified architecture. Designed to generate high-quality video up to 720p on a single GPU, SANA-Video 2.0 matches full-softmax video DiTs in quality while retaining the favo…
Haotian Liang, Mingkang Chen, Yufei Huang, Yuchun Guo, Xiaomeng Zhu, Xiangli Shi, et al.
Embodied cognition requires agents to connect high-level task reasoning with the physical states to be achieved. We introduce Hy-Embodied-RxBrain, an embodied cognition foundation model with joint language-visual reasoning and imagination. Unlike vision-language models that empha…
Lirui Zhao, Modi Shi, Li Chen, Qi Liu, Ping Luo, Hongyang Li
Modern robot learning systems increasingly rely on dense progress or value signals to evaluate intermediate states, guide policy learning, and detect task completion, making the quality of these signals critical. Since such dense labels are rarely available at scale, normalized t…
Tianxing Chen, Yue Chen, Zixuan Li, Junyuan Tang, Kailun Su, Haoran Lu, et al.
Generalist robot manipulation policies have advanced rapidly, yet existing benchmarks remain limited in systematically evaluating their capabilities. Many rely on simple, short-horizon, or skill-narrow tasks with limited capability coverage, and are often conducted only in simula…
Bowen Jing, Mingxin Wang, Ruiyang Hao, Chenchen Ge, Hanwen Shen, Junjie He, et al.
Deformable object manipulation poses challenges beyond task completion: successful execution must also maintain safe physical interaction, holding the object stably without slip or drop while avoiding excessive deformation. However, existing manipulation benchmarks are predominan…
Shuai Li, Qinglin Wang, Ping Luo, Jiahuan Wang, Hongyang Hu, Haotian Mo, et al.
Federated Transformer training increasingly relies on local AdamW, whose adaptive updates can provide much stronger local progress than SGD-based training. However, under heterogeneous client data, even globally corrected AdamW updates may remain highly uneven in coordinate-wise…