In unsupervised person Re-ID, peer-teaching strategy leveraging two networks to facilitate training has been proven to be an effective method to deal with the pseudo label noise. However, training two networks with a set of noisy pseudo labels reduces the complementarity of the two networks and results in label noise accumulation. To handle this issue, this paper proposes a novel Dual Clustering Co-teaching (DCCT) approach. DCCT mainly exploits the features extracted by two networks to generate two sets of pseudo labels separately by clustering with different parameters. Each network is trained with the pseudo labels generated by its peer network, which can increase the complementarity of the two networks to reduce the impact of noises. Furthermore, we propose dual clustering with dynamic parameters (DCDP) to make the network adaptive and robust to dynamically changing clustering parameters. Moreover, Consistent Sample Mining (CSM) is proposed to find the samples with unchanged pseudo labels during training for potential noisy sample removal. Extensive experiments demonstrate the effectiveness of the proposed method, which outperforms the state-of-the-art unsupervised person Re-ID methods by a considerable margin and surpasses most methods utilizing camera information.

In the tasks of multi-robot collaborative area search, we propose the unified approach for simultaneous mapping for sensing more targets (exploration) while searching and locating the targets (coverage). Specifically, we implement a hierarchical multi-agent reinforcement learning algorithm to decouple task planning from task execution. The role concept is integrated into the upper-level task planning for role selection, which enables robots to learn the role based on the state status from the upper-view. Besides, an intelligent role switching mechanism enables the role selection module to function between two timesteps, promoting both exploration and coverage interchangeably. Then, the primitive policy learns how to plan based on their assigned roles and local observation for sub-task execution. The well-designed experiments show the scalability and generalization of our method compared with state-of-the-art approaches in scenes with varying complexity and numbers of robots. Our code is released at https://github.com/linaug/Role_selection Note to Practitioners—In the realm of area search tasks, the collaboration among multiple robots in a multi-robot system brings about a reduction in labor costs and an enhancement in task completion efficiency. Recognizing this, the development of an intelligent multi-robot collaborative decision-making algorithm becomes crucial-especially within complex and diverse environments, ensuring better adaptability to various real-world scenarios. This paper introduces a hierarchical framework designed for the simultaneous mapping and searching of targets in multi-robot area search tasks. The upper-level role selection framework employs adaptive role learning through deep reinforcement learning, while the low-level primitive policy focuses on planning actions based on the designated roles. As a result, the proposed approach exhibits notable advantages in terms of both the percentage and efficiency of task completion, particularly evident in well-designed scenes with varying complexities. We believe this framework holds significant potential for real-world applications, addressing challenges in multi-robot collaboration across diverse environments.

This paper mainly studies the deployment of hybrid sensors on highways. By constructing the deployment location constraint model, the overall accuracy of sensor deployment can be maximized. In addition, in order to meet the needs of intelligent transportation, the consideration of traffic data communication is added to the work. The highway under study is first divided into several units, and the combination type of sensors is used to represent the possible layout of two adjacent sensors. Then, a 0–1 optimization model reflecting the interaction between the sensor position and the server position is established. Then, a two-step search algorithm is proposed to find the optimal solution of the model and determine the deployment scheme with maximum accuracy. Finally, an example is given to verify the method. The results show that there are significant differences between uniform unit-deployment schemes and non-uniform unit-deployment schemes. Through the sensitivity analysis of each factor, the influence of budget and communication radius on the deployment plan is proven. In addition, the ramp length can also have a negative impact on the target value.

3D vehicle detection still remains challenging due to the occlusion, illumination and special weather. This paper proposes a data-level fusion method to combine the images and LiDAR to achieve 3D vehicle detection in the foggy environment. This method first restores the corrupted multi-modal data though image defogging and point cloud defogging. Then the pseudo-LiDAR is generated by back-projecting the image pixels with the estimated image depth. After revisitation and transformation, the pseudo-LiDAR and LiDAR data can be fused in the 3D space. Finally, the fused point clouds are fed into the BtcDet detection network to obtain detection results. Furthermore, in order to evaluate 3D detection methods, this paper establishes a multi-modal dataset (images and LiDAR) of the foggy scenarios by simulation algorithms to compensate for the lack of foggy dataset. Based on the synthetic dataset, the experiments are conducted to evaluate the proposed methods and other comparative methods. The accuracy of the 3D target detection algorithm is higher than comparative methods.