In order to solve the adaptation problem of dynamic change of cargo throughput and the problem of heterogeneous equipment collaboration in large-scale logistics sorting scenarios, this paper proposes a multi-agent heterogeneous hybrid safe sorting system based on Transformer. By constructing the CTCE/CTDE hybrid multi-agent reinforcement learning framework, the collaborative optimization of centralized task scheduling and distributed path planning is realized. The system uses the Transformer architecture to process the global logistics state, innovatively sets the task state space to realize the task quantization in network training, combines the MAPPO algorithm to realize the real-time obstacle avoidance of UAV and UGV, and introduces a safety projection mechanism to correct abnormal actions to the feasible domain in real time. This research provides a solution with both efficiency and safety for automated sorting in a dynamic logistics environment.

To address the challenge of low computational efficiency in UAV navigation within dynamic logistics warehouse environments, this paper presents a deep reinforcement learning (DRL) framework based on unified obstacle state representation. By integrating static and dynamic obstacles into a single "obstacle state" vector and eliminating the traditional voxel map maintenance mechanism, the proposed method significantly reduces computational complexity. The framework models the navigation task as a Markov Decision Process (MDP), using a lightweight neural network with Proximal Policy Optimization (PPO) to achieve real-time obstacle avoidance. Experimental results in simulated high-dynamic environments show a 95% collision-free success rate. This framework is particularly suitable for resource-constrained scenarios such as indoor logistics, where real-time adaptability to dynamic obstacles is critical.

Traditional discriminative correlation filter (DCF) has received great popularity due to its high computational efficiency. However, the lightweight framework of DCF cannot promise robust performance when the tracker faces appearance variations within the background. These unpredictable appearance variations always distract the filter. Most existing DCF-based trackers either utilize deep convolutional features or incorporate additional constraints to elevate tracking robustness. Despite some improvements, both of them hamper the tracking speed and can only roughly alleviate the distractions of appearance variations. In this paper, a novel spatial reliability enhanced learning strategy is proposed to handle the problems aforementioned. By monitoring the variation of response produced in detection phase, a dynamic reliability map is generated to indicate the reliability of each background subregion. Then, label adjustment is conducted to repress the distractions of these unreliable areas. Compared with the conventional way of constraint where a new term is always added to realize the desired goal, label adjustment is simultaneously more efficient and effective. Moreover, to promise the accuracy and dependability of the reliability map, an adaptively updated response pool recording reliable historical response values is proposed. Extensive and exhaustive experiments on three challenging unmanned aerial vehicle (UAV) benchmarks, i.e., UAV123@10fps, DTB70 and UAVDT, which totally include 243 video sequences, validate the superiority of the proposed method against other state-of-the-art trackers and exhibit a remarkable generality in a variety of scenarios. Meanwhile, the tracking speed of 65.2FPS on a cheap CPU makes it suitable for real-time UAV applications.

In many industrial applications of unmanned aerial manipulator (UAM), visual approaching the object is crucial to subsequent manipulating. In comparison with the widely-studied manipulating, the key to efficient vision-based UAM approaching, i.e., UAM object tracking, is still limited. Since traditional model-based UAM tracking is costly and cannot track arbitrary objects, an intuitive solution is to introduce state-of-the-art model-free Siamese trackers from the visual tracking field. Although Siamese tracking is most suitable for the onboard embedded processors, severe object scale variation in UAM tracking brings formidable challenges. To address these problems, this work proposes a novel model-free scale-aware Siamese tracker (SiamSA). Specifically, a scale attention network is proposed to emphasize scale awareness in feature processing. A scale-aware anchor proposal network is designed to achieve anchor proposing. Besides, two novel UAM tracking benchmarks are first recorded. Comprehensive experiments on benchmarks validate the effectiveness of SiamSA. Furthermore, real-world tests also confirm practicality for industrial UAM approaching tasks with high efficiency and robustness.

Although the manipulating of the unmanned aerial manipulator (UAM) has been widely studied, vision-based UAM approaching, which is crucial to the subsequent manipulating, generally lacks effective design. The key to the visual UAM approaching lies in object tracking, while current UAM tracking typically relies on costly model-based methods. Besides, UAM approaching often confronts more severe object scale variation issues, which makes it inappro-priate to directly employ state-of-the-art model-free Siamese-based methods from the object tracking field. To address the above problems, this work proposes a novel Siamese network with pairwise scale-channel attention (SiamSA) for vision-based UAM approaching. Specifically, SiamSA consists of a pairwise scale-channel attention network (PSAN) and a scale-aware anchor proposal network (SA-APN). PSAN acquires valuable scale information for feature processing, while SA-APN mainly attaches scale awareness to anchor proposing. Moreover, a new tracking benchmark for UAM approaching, namely UAMT100, is recorded with 35K frames on a flying UAM platform for evaluation. Exhaustive experiments on the benchmarks and real-world tests validate the efficiency and practicality of SiamSA with a promising speed. Both the code and UAMT100 benchmark are now available at https://github.com/vision4robotics/SiamSA.

Most existing trackers based on discriminative correlation filters (DCF) try to introduce predefined regularization term to improve the learning of target objects, e.g., by suppressing background learning or by restricting change rate of correlation filters. However, predefined parameters introduce much effort in tuning them and they still fail to adapt to new situations that the designer did not think of. In this work, a novel approach is proposed to online automatically and adaptively learn spatio-temporal regularization term. Spatially local response map variation is introduced as spatial regularization to make DCF focus on the learning of trust-worthy parts of the object, and global response map variation determines the updating rate of the filter. Extensive experiments on four UAV benchmarks have proven the superiority of our method compared to the state-of-the-art CPU- and GPU-based trackers, with a speed of ~60 frames per second running on a single CPU. Our tracker is additionally proposed to be applied in UAV localization. Considerable tests in the indoor practical scenarios have proven the effectiveness and versatility of our localization method. The code is available at https://github.com/vision4robotics/AutoTrack.

A reliable pose estimation is crucial in the landing tasks of helicopters. This paper mainly focuses on presenting a smooth and reliable pose estimation method during helicopter landing using Augmented Reality (AR) markers. Based on the proposed hierarchical fiducial marker system called Hierarchical Augmented Reality Code (HArCo), pose estimation can be performed within a much longer range. The design of the system, generation of the marker dictionary and the application in helicopter landing are thoroughly described in this paper. The performance of the pose estimation algorithm based on HArCo is tested in a landing task. As pose information is accessible throughout the landing procedure, a safe auto-landing based on HArCo can also be conducted.

Output containment control problems for high-order linear time-invariant swarm systems are investigated. Firstly, output containment protocols are presented for leaders and followers respectively to make sure that the output dynamics property of leaders can be improved and the outputs of followers can converge to the convex hull formed by the outputs of leaders. Then output containment problems for swarm systems are transformed into stability problems, and sufficient conditions for swarm systems to achieve output containment are proposed. Moreover, an approach to determine the gain matrix in the output containment protocol is given, which has less calculation complexity. Finally, numerical simulations are presented to demonstrate theoretical results.