RefGuard: Identity-Aware Language-Guided Robot Manipulation via Joint Target-Anchor-Frame Grounding
为解决机器人在复杂场景中因对象识别错误导致的操作失误,提出RefGuard框架,通过联合目标-锚点-参照系定位方法提高操作准确性。
为解决机器人在复杂场景中因对象识别错误导致的操作失误,提出RefGuard框架,通过联合目标-锚点-参照系定位方法提高操作准确性。
本文通过在MAST-U PCS中集成基于神经网络的虚拟电路,实现等离子体形状的实时控制,并强调了验证流程以确保控制框架的可靠性。
This study addresses the challenge of lateral vibration suppression in cooperative transportation of large flexible payloads by heterogeneous robots. A passive force-control collaborative framework is proposed, employing a leader-follower architecture that integrates velocity command shaping with admittance control to establish an equivalent mass-spring-damper model. The system’s energy dissipation stability is rigorously proven through passivity analysis. Both simulations and experimental results demonstrate that this strategy effectively achieves passive vibration damping and stable cooperative transport for heavy flexible loads. Consequently, the proposed method successfully resolves critical issues regarding compliant control and stability assurance in heterogeneous robotic collaboration, offering a robust solution for manipulating large-scale flexible objects without active feedback complexity.
This work addresses the limitations of conventional virtual circuits in tokamak plasma shape control, which rely on offline precomputation and struggle to adapt to real-time dynamic deviations, leading to degraded performance. The authors propose a deep neural network–based plasma shape parameter simulator trained on a Grad–Shafranov equilibrium database comprising over one million samples. This approach enables, for the first time, differentiable and highly accurate real-time generation of virtual circuits. By dynamically producing orthogonal virtual circuits conditioned on the current plasma state, the method overcomes the constraints of traditional offline scheduling and effectively decouples multivariable control. Experimental validation demonstrates that the generated circuits maintain high fidelity and strong decoupling across diverse equilibrium configurations, significantly enhancing the robustness and generalizability of real-time plasma shape control.
This work proposes TokaMind, a multimodal Transformer-based pre-trained foundation model designed to address key challenges in tokamak plasma modeling—namely, the heterogeneity of multimodal diagnostic data, inconsistent sampling rates, and missing signals. TokaMind introduces a novel training-free DCT3D embedding method that enables unified representation of diverse data modalities, including time series, 2D profiles, and video streams. The architecture features plug-and-play embedding interfaces and component-wise selective loading, facilitating efficient fine-tuning and transfer learning. Integrated with VAE-based surrogate embeddings and robust mechanisms for handling missing signals, TokaMind significantly outperforms baseline models on the TokaMark benchmark using the MAST dataset. Notably, lightweight fine-tuning of the pre-trained model surpasses performance achieved by training from scratch, demonstrating the efficacy and generalization capability of multimodal pre-training in fusion plasma diagnostics.
为解决机器人在复杂场景中因对象识别错误导致的操作失误,提出RefGuard框架,通过联合目标-锚点-参照系定位方法提高操作准确性。
本文通过在MAST-U PCS中集成基于神经网络的虚拟电路,实现等离子体形状的实时控制,并强调了验证流程以确保控制框架的可靠性。
This study addresses the challenge of lateral vibration suppression in cooperative transportation of large flexible payloads by heterogeneous robots. A passive force-control collaborative framework is proposed, employing a leader-follower architecture that integrates velocity command shaping with admittance control to establish an equivalent mass-spring-damper model. The system’s energy dissipation stability is rigorously proven through passivity analysis. Both simulations and experimental results demonstrate that this strategy effectively achieves passive vibration damping and stable cooperative transport for heavy flexible loads. Consequently, the proposed method successfully resolves critical issues regarding compliant control and stability assurance in heterogeneous robotic collaboration, offering a robust solution for manipulating large-scale flexible objects without active feedback complexity.
This work addresses the limitations of conventional virtual circuits in tokamak plasma shape control, which rely on offline precomputation and struggle to adapt to real-time dynamic deviations, leading to degraded performance. The authors propose a deep neural network–based plasma shape parameter simulator trained on a Grad–Shafranov equilibrium database comprising over one million samples. This approach enables, for the first time, differentiable and highly accurate real-time generation of virtual circuits. By dynamically producing orthogonal virtual circuits conditioned on the current plasma state, the method overcomes the constraints of traditional offline scheduling and effectively decouples multivariable control. Experimental validation demonstrates that the generated circuits maintain high fidelity and strong decoupling across diverse equilibrium configurations, significantly enhancing the robustness and generalizability of real-time plasma shape control.
This work proposes TokaMind, a multimodal Transformer-based pre-trained foundation model designed to address key challenges in tokamak plasma modeling—namely, the heterogeneity of multimodal diagnostic data, inconsistent sampling rates, and missing signals. TokaMind introduces a novel training-free DCT3D embedding method that enables unified representation of diverse data modalities, including time series, 2D profiles, and video streams. The architecture features plug-and-play embedding interfaces and component-wise selective loading, facilitating efficient fine-tuning and transfer learning. Integrated with VAE-based surrogate embeddings and robust mechanisms for handling missing signals, TokaMind significantly outperforms baseline models on the TokaMark benchmark using the MAST dataset. Notably, lightweight fine-tuning of the pre-trained model surpasses performance achieved by training from scratch, demonstrating the efficacy and generalization capability of multimodal pre-training in fusion plasma diagnostics.