Chengdong Wu

This paper presents the Post-Stimulus Time-Dependent Event Descriptor (P-TED), a novel “pure event” feature descriptor designed for neuromorphic vision data. Unlike conventional frame-based approaches or hybrid methods that transform event data into intermediate representations, P-TED operates directly on asynchronous event streams, thereby preserving the intrinsic low-latency and high-temporal-resolution advantages of event-based sensors. The descriptor integrates two complementary feature sets: a motion feature vector, which aggregates spatial relationships within a Moore neighbourhood to quantify stimulus direction, and a pattern feature vector, which employs rate encoding to capture temporal excitation signatures. The efficacy of the P-TED framework is validated through three distinct experiments: object and character recognition (MNIST-DVS and CIFAR10-DVS), mobile robot movement analysis, and complex non-rigid robotic hand gesture recognition (RoShamBo). Experimental results demonstrate that the P-TED achieves a significant reduction in classification latency, requiring only 2.7 ms compared to the 10.3 ms recorded by the state-of-the-art Distribution-Aware Retinal Transform (DART) framework. Additionally, P-TED exhibits superior robustness in disambiguating symmetric and mirrored motions, as well as in maintaining stability under non-linear fluctuations in event density caused by changing scale. This work establishes P-TED as a high-speed, computationally efficient, and explainable solution for real-time neuromorphic robotic vision systems.

Dynamic underwater novel-view synthesis remains challenging because refraction, scattering, and particle interference undermine correspondence reliability, geometric consistency, and appearance stability. We propose SeaAnchor-GS, a robust dynamic 3D Gaussian splatting framework for underwater scene reconstruction. Each Gaussian is associated with a persistent identity embedding and deformed via identity–time conditioning, which improves deformation estimation under unstable underwater observations. A dual-branch residual dynamics module captures both dominant motion and fine-scale variations, while confidence-guided sampling, progressive deformation activation, and neighborhood-consistency regularization enhance optimization robustness and model compactness. Extensive experiments on dynamic underwater benchmarks show consistent improvements in reconstruction fidelity and perceptual quality, with a favorable balance between quality, efficiency, and representation compactness. Additional results on a static underwater benchmark suggest that the proposed representation remains competitive beyond the dynamic setting.

Aiming at the problems of difficult hard-constraint enforcement and weak environmental generalization ability in the safe trajectory planning of manipulators in complex environments, a Policy-Guided Model Predictive Path Integral (PG-MPPI) planning framework is proposed. This framework integrates the advantages of reinforcement learning and model predictive control to construct a global prior guidance, local real-time optimization and hard-constraint safety assurance: a Constraint-Discounted Soft Actor–Critic (CD-SAC) offline learning policy is designed, which incorporates the configuration-space distance field as a safety guidance term to realize the learning of obstacle avoidance behavior; the offline policy is used to guide the online sampling and optimization of MPPI, improving sampling efficiency and planning quality; and a Control Barrier Function (CBF) safety filter is introduced to revise control commands in real time, ensuring the strict satisfaction of constraints. Taking the SIASUN T12B manipulator as the research object, simulation comparison experiments are carried out in multi-obstacle scenarios. The results show that the PG-MPPI algorithm outperforms the comparison algorithms in the success rate of collision-free target reaching, ensures the smoothness and feasibility of the trajectory, and has a good adaptive capacity to complex environments with unknown obstacle configurations, thus providing an efficient solution for the autonomous and safe operation of manipulators.

Aiming at the problems of difficult hard constraint enforcement, weak environmental generalization ability in the safe trajectory planning of manipulators in complex environments, a Policy-Guided Model Predictive Path Integral (PG-MPPI) planning framework is proposed. This framework integrates the advantages of reinforcement learning and model predictive control to construct a global prior guidance, local real-time optimization and hard constraint safety assurance: a Constraint-Discounted Soft Actor-Critic (CD-SAC) offline learning policy is designed, which incorporates the configuration-space distance field as a safety guidance term to realize the learning of obstacle avoidance behavior; the offline policy is used to guide the online sampling and optimization of MPPI, improving sampling efficiency and planning quality; a Control Barrier Function (CBF) safety filter is introduced to revise control commands in real time, ensuring the strict satisfaction of constraints. Taking the SIASUN T12B manipulator as the research object, simulation comparison experiments are carried out in multi-obstacle scenarios. The results show that the PG-MPPI algorithm outperforms the comparison algorithms in the success rate of collision-free target reaching, ensure the smoothness and feasibility of the trajectory, and has a good adaptive capacity to dynamic environments, thus providing an efficient solution for the autonomous and safe operation of manipulators.

Synthetic Aperture Radar (SAR) and optical image matching is a critical prerequisite for achieving image fusion. Due to nonlinear radiometric differences and geometric distortions between SAR and optical images, existing methods struggle to extract homologous feature points, resulting in suboptimal registration accuracy. To address the mismatch issues in SAR-optical cross-modal image matching, a feature-based matching method named HFTM is proposed in this letter. Firstly, a hierarchical feature extraction network is applied on multimodal image pairs to extract global and local context information to obtain robust feature points. Consistent feature matching is then achieved in the local region through a Transformer-based attention mechanism and NeighConsensus (TNC) Matching network. The results of local matching are fed into the proposed global Position-encoded Probabilistic (P2) Matching network, which employs a detailed matching layer to generate an enhanced feature map, and predicts robust matching point pairs by a spatial expectation operator. Experimental results demonstrate the superior performance of our method, achieving a single responsiveness estimation accuracy of 89.43% on QXS-SAROPT and 86.21% on SEN12MS at a 5-pixel threshold.