Action supervision in vision-language-action (VLA) models is often treated as a downstream objective for learning action prediction. In this paper, we study it instead as a force that shapes inherited multimodal representations. We show that this shaping has a dual effect: it is necessary for forming action-compatible representations, but when action supervision is applied too directly to the inherited multimodal pathway, it can also destabilize representations that support language-side processing and object grounding. To address this tension, we introduce Action QFormer, a query-based action-facing interface that uses instruction-conditioned queries to reorganize inherited multimodal information into action-facing representations before downstream action generation. In zero-shot sim-to-real navigation, Action QFormer improves average closed-loop task success from 18.8% to 56.3%, raises fixed-instruction action-generation correctness from 22.5% to 75.5%, and nearly eliminates out-of-distribution instruction generations. Further analyses show that Action QFormer changes how action supervision shapes inherited multimodal representations, reducing broad upstream rewriting while preserving targeted and sometimes constructive action-supervised adaptation. These results suggest that improving VLA performance requires not only stronger pretrained backbones, but also better ways of selecting and organizing inherited multimodal information while controlling how it is shaped under action supervision.
Yufeng Ji, Wenhao Tang, Haoyi Niu et al.· 0 citations
We introduce trajectory-regularized stochastic optimal control (TRSOC), which augments standard stochastic optimal control (SOC) with a Kullback--Leibler (KL) divergence between controlled and reference trajectory distributions. Using Girsanov's theorem, the trajectory KL reduces to a quadratic drift mismatch penalty, yielding a modified running cost that preserves the dynamic programming (DP) structure. We derive the corresponding Hamilton--Jacobi--Bellman (HJB) equation and characterize the optimal policy. In the linear-quadratic (LQ) setting, the formulation admits a closed-form solution with an augmented control cost. Experiments show that the regularization parameter induces a trade-off between performance-driven and reference-preserving behavior, including cases with reference dynamics learned from offline data.
Humanoid robots stand at the forefront of robotics, aiming to capture the agility, robustness, and expressivity of human movement in an anthropomorphic form. The locomotion control of humanoids has evolved from classical model-based methods to reinforcement learning powered by large-scale simulation and now to generative models that produce adaptive, whole-body behaviors, propelling humanoids toward operation in real-world environments. This survey positions humanoid control at a turning point, converging toward a unified paradigm of physics-guided generative intelligence that integrates optimization, learning, and predictive reasoning. We identify three core principles linking these paradigms: physics-based modeling, constrained decision-making, and adaptation to uncertainty. Building on these connections, we provide recommendations for researchers and outline open challenges in safety, accessibility, and human-level capability. These directions represent a transformation from engineered stability to intelligent autonomy, laying the groundwork for humanoid generalists capable of operating safely, collaborating naturally, and extending human capability in the open world.
Yan Gu, Guanya Shi, Fan Shi et al.· Science Robotics· 7 citations