Spatio-temporal tasks often encounter incomplete data arising from missing or inaccessible sensors, making spatio-temporal kriging crucial for inferring the completely missing temporal information. However, current models struggle with ensuring the validity and generalizability of inferred spatio-temporal patterns, especially in capturing dynamic spatial dependencies and temporal shifts, and optimizing the generalizability of unknown sensors. To overcome these limitations, we propose Spatio-Temporal Aware Graph Adversarial Neural Network (STA-GANN), a novel GNN-based kriging framework that improves spatio-temporal pattern validity and generalization. STA-GANN integrates (i) Decoupled Phase Module that senses and adjusts for timestamp shifts. (ii) Dynamic Data-Driven Metadata Graph Modeling to update spatial relationships using temporal data and metadata; (iii) An adversarial transfer learning strategy to ensure generalizability. Extensive validation across nine datasets from four fields and theoretical evidence both demonstrate the superior performance of STA-GANN.

Visual perception is an important component for autonomous navigation of unmanned surface vessels (USV), particularly for the tasks related to autonomous inspection and tracking. These tasks involve vision-based navigation techniques to identify the target for navigation. Reduced visibility under extreme weather conditions in marine environments makes it difficult for vision-based approaches to work properly. To overcome these issues, this paper presents an autonomous vision-based navigation framework for tracking target objects in extreme marine conditions. The proposed framework consists of an integrated perception pipeline that uses a generative adversarial network (GAN) to remove noise and highlight the object features before passing them to the object detector (i.e., YOLOv5). The detected visual features are then used by the USV to track the target. The proposed framework has been thoroughly tested in simulation under extremely reduced visibility due to sandstorms and fog. The results are compared with state-of-the-art de-hazing methods across the benchmarked MBZIRC simulation dataset, on which the proposed scheme has outperformed the existing methods across various metrics.

Conventional multi-sensor multi-object tracking algorithms usually fuse local positions to construct the global situation. However, low-cost cameras that have been widely-used in small-scale UAVs only provide bearing angle measurement but not target positions, which prohibits the application of conventional tracking paradigms. We propose a solution of vision based multi-object tracking through UAV swarm. Given the videos captured by UAVs and the states of the UAVs, the proposed solution fuses visual and geometry information to tackle three tasks: (1) associating the targets reported by different UAVs. (2) computing the targets’ positions in inertial coordinate system. (3) associating the targets reported at different instants. The effectiveness of the proposed solution is evaluated by offline ablation experiments, field scene experiments, and online closed-loop simulation. The source code is available at https://github.com/bitshenwenxiao/MOTL and the simulation video is available at https://youtu.be/erCiENAOEaM.

This paper studies the problem of air-to-air target following of micro aerial vehicles (MAVs) motivated by the application of defense against malicious MAVs. When the bearing of the target MAV has been measured by the onboard visual sensor of the pursuer MAV, the problem becomes three-dimensional (3-D) bearing-only target following, which has been rarely studied in the literature and faces some unique challenges. To solve this problem, we propose the following novel results. First, to estimate the motion of the target MAV from 3-D bearing measurements, we propose a new pseudo-linear Kalman filter, which has a concise expression and superior stability compared to the classic ones such as the extended Kalman filter and modified polar coordinate filter. Second, we propose a novel approach to analyze the observability of state estimation when only bearing information is available. While the existing approaches are applicable to 2-D and single-step time-horizon cases, ours can handle more general 3-D and multiple-step time-horizon cases. Third, based on the theoretical conclusion of our observability analysis, we design a new 3-D helical guidance law that can better exploit the additional degree of freedom in 3D. The guidance law is adapted to the quadcopter's dynamics and a low-level flight controller is designed based on geometric control. Numerical simulation results verify the superior performance of the proposed algorithms compared to the state-of-the-art ones. Flight experiments on real quadrotor platforms further show the effectiveness and robustness of the proposed algorithms in practice.

This paper investigates the problem of one-step ahead optimal active sensing strategy to minimise estimation errors with range-only measurements for non-manoeuvring target. The determinant of Fisher Information Matrix (FIM) is utilized as the objective function in the proposed optimisation problem since it quantifies the volume of uncertainty ellipsoid of any efficient estimator. In consideration of physical velocity and turning rate constraints, the optimal heading angle command that maximises the cost function is derived analytically. Simulations are conducted to validate the analytical findings.

This paper addresses the problem of maneuver recognition and behavior anomaly detection for generic targets by means of pattern matching techniques. The problem analysis is performed making specific reference to moving vehicles in a multi-lane road scenario, but the proposed technique can be easily extended to significantly different monitoring contexts. The potential extensions include, but are not limited to, public surveillance in train station or airport, road incidents and relative precursors detection, and vehicle trajectories monitoring. The overall proposed solution consists of a trajectory analysis tool and a string-matching method. This allows the integration of two different approaches, to detect both a priori defined patterns of interest and generic maneuver/behavior standing out from those regularly exhibited. The proposed string matching algorithm is newly developed in this paper, based on Regular Expressions. For generating reference patterns, a technique for the automatic definition of a dictionary of regular expressions matching the commonly observed target maneuvers is developed. The advantages of the proposed approach are extensively analyzed and tested by means of numerical simulations and experiments.

This paper considers the problem of two-dimensional constrained trajectory optimization of a point-mass aerial robot for constant-maneuvering target localization using bearing-only measurement. A performance metric that can be utilized in trajectory optimization to maximize target observability is proposed first based on geometric conditions. One-step optimal maneuver that maximizes the observability criterion is then derived analytically for moving targets. The heading angle constraint is also incorporated in the proposed optimal maneuver derivation to support practical application. Numerical simulations with some comparisons are presented to validate the analytical findings.