Diffusion and flow models provide expressive policy classes for robotic and sequential decision-making systems. However, their action likelihoods are generally intractable, limiting the direct use of widely adopted policy optimization methods such as PPO and GRPO. We propose Doob Policy Optimization (DoobPO), a framewo...
Mao-Jiang Su, Qi-Jie Zhu, Yu-Shuang Xiao et al.· 0 citations
PairFormer-based all-atom models demonstrate strong capabilities in biomolecular binder design. However, existing methods often use all-atom representations for modality-specific binder design, such as protein and peptide design. In contrast, recent geometric latent diffusion models pursue unified binder design by repr...
Mao-Jiang Su, Jia-Le Chen, Qi-Jie Zhu et al.· 0 citations
Embodied agents now take on ever longer tasks. For long tasks, knowing only whether a task finally succeeds or fails says little; the steps along the way matter. Progress Reward Models (PRMs) score how far a task has come at every step, and serve as dense rewards, verifiers and monitors. Yet in long tasks the current f...
Jian-Shu Zhang, Ke-Liang Wu, Cheng-Xuan Qian et al.· 0 citations
This work introduces Video2Skill, a benchmark that covers robot tabletop manipulation and human kitchen activity and tests three core capabilities: locating manipulation events in time, grouping events of the same transformation, and deciding when to reuse an existing skill or create a new one.
Jian-Shu Zhang, Ce Zhang, Xi-Yuan Yang et al.· 0 citations
Under the evaluated four-image setting, these results indicate that separator-based long-sequence pretraining improves recognition robustness and dense visual prediction relative to ARM.
Han Liu, Zi-Dan Wang, Shuo-Xi Zhang et al.· arXiv.org· 0 citations
A unified view of progress reward modeling for robotic learning is provided in three connected steps that connect what a progress model is, how it is built, and how its quality is validated.
Jian-Shu Zhang, Ke-Liang Wu, Haoran Lu et al.· arXiv.org· 5 citations
APT-Tune is proposed, a parameter-efficient recipe that teaches VLMs to use causal transitions without forgetting how to answer video questions, and substantially improves APT recall while also improving event-level video transfer.
Shang Wu, Haoran Lu, Song-Lin Liu et al.· arXiv.org· 0 citations
From YAML-first specifications that decouple contents, placement, behavior, and agent exposure, MagicSim constructs diverse executable worlds spanning task families, interaction regimes, physics, layouts, sensors, avatars, and robot embodiments in one reset-and-step loop.