Sage Cao
@sagecao.bsky.social
Phd student at HKU IDS, exploring robot learning, generative models, and cognitive-inspired models
BS in Math @SYSU, MPhil @HKUST-GZ
Personal Website: https://sagecao1125.github.io/
BS in Math @SYSU, MPhil @HKUST-GZ
Personal Website: https://sagecao1125.github.io/
Surprisingly simple: combine the scores of existing diffusion/flow policies and get stronger control—no retraining, no new data. That’s our General Policy Composition (GPC).
arxiv.org/pdf/2510.01068
Works across VA & VLA, different modalities. Theory + sim/real robots back it up.
#Robotics #AI #ML
arxiv.org/pdf/2510.01068
Works across VA & VLA, different modalities. Theory + sim/real robots back it up.
#Robotics #AI #ML
October 6, 2025 at 3:48 PM
Surprisingly simple: combine the scores of existing diffusion/flow policies and get stronger control—no retraining, no new data. That’s our General Policy Composition (GPC).
arxiv.org/pdf/2510.01068
Works across VA & VLA, different modalities. Theory + sim/real robots back it up.
#Robotics #AI #ML
arxiv.org/pdf/2510.01068
Works across VA & VLA, different modalities. Theory + sim/real robots back it up.
#Robotics #AI #ML