Forecasting vehicular motions in autonomous driving requires a deep understanding of agent interactions and the preservation of motion equivariance under Euclidean geometric transformations. Traditional models often lack the sophistication needed to handle the intricate dynamics inherent to autonomous vehicles and the interaction relationships among agents in the scene. As a result, these models have a lower model capacity, which then leads to higher prediction errors and lower training efficiency. In our research, we employ EqMotion, a leading equivariant particle, and human prediction model that also accounts for invariant agent interactions, for the task of multi-agent vehicle motion forecasting. In addition, we use a multi-modal prediction mechanism to account for multiple possible future paths in a probabilistic manner. By leveraging EqMotion, our model achieves state-of-the-art (SOTA) performance with fewer parameters (1.2 million) and a significantly reduced training time (less than 2 hours).
Although the use of smart phones in working environment is ubiquitous, the research on how the use of smart phones affects the working results is still in its infancy. Based on social learning theory, expectancy violation theory and resource conservation theory, this study explores how the boss phubbing (Bphubbing) affects employees' performance. Through the survey and data analysis of 215 employees, the results show that: (1) There is a negative correlation between Bphubbing and employee job performance. (2) Employees ‘ psychological capital plays a full mediation role in the relationship between Bphubbing and employee job performance. (3) work-family enrichment has a negative moderating effect on the relationship between Bphubbing and employees ’ psychological capital, which confirms the reverse stress-buffering model. This study expands the research on the mechanism of the negative impact of improper use of smart phones on the organization, which is conducive to the organization and managers to scientifically view the positive and negative sides of the use of smart phones for information management in the workplace, so as to put forward scientific suggestions for the correct use of smart phones.
Humanoid robots show the ability to replace humans for some dangerous tasks in complex environment. There are various motion types for robots moving into the dangerous environment like walking, rolling and crawling. However, the higher centroid brings instability when humanoid robots execute walking motion. Rolling motion may cause damage crash to the robot's mechanical structure. Therefore, a crawling action with low-centroid and less collision force is designed for robots to execute dangerous tasks. Firstly, a mechanical structure is designed for our robot called BHR-FCR. BHR-FCR own 23-Degree of Freedom (DoF) which provides possibility to achieve crawling motion. Then, Newton-Euler iterative recursion and Lagrange method are adopted for dynamics analysis which provides the basis for trajectory planning. Moreover, in the view of energy loss, the whole-body motion trajectory is generated by dynamic model and trajectory optimization with collocation method. The simulation and real-world experiments depict that BHR-FCR robot is able to crawl across lower wall and crawl on the slope which illustrates the stability of BHR-FCR mechanical structure and the effectiveness proposed algorithm.
Due to the robustness of ConvNet landmarks to appearance changes and viewpoint changes, there are some excellent works in ConvNet landmarks based visual place recognition (VPR). However, when dealing with extremely complex environmental changes, the performance of ConvNet landmarks based VPR is also difficult to meet the needs of practical applications because of the lack of information. This paper proposes a robust visual place recognition in challenging environments using landmarks associated with spatiotemporal and semantic Information (LSTS-VPR). LSTS-VPR combines ConvNet landmarks and image sequence to take advantage of the rich information of the objects contained in the sequence at different views and different time periods to filter and obtain more discriminative landmarks such as buildings and trees. LSTS-VPR consists of two modules, namely landmarks associated with spatiotemporal and semantic information-based feature aggregation module (LSTS-FAM) and multi-stage filtering-based feature matching module (MF-FMM). In LSTS-FAM, we use tube linking algorithm to connect the similar landmarks in different frames of one sequence and then calculate the landmark weights to screen the landmarks with high repeatability, which are proportional to the length of the tube. In MF-FMM, we use the semantic similarity, center distance and geometric proportion of landmarks as constraints to obtain the best matching sequence more accurately. Finally, the results of experiments on eight datasets demonstrate that the method proposed in this paper can better cope with a variety of changes including appearance changes, viewpoint changes, and occlusion of dynamic targets than state-of -art VPR approaches.
In this paper, we report on our use of cloud-robotics solutions to teach a Robotics Applications Programming course at Zurich University of Applied Sciences (ZHAW). The usage of Kubernetes based cloud computing environment combined with real robots - turtlebots and Niryo arms - allowed us to: 1) minimize the set up times required to provide a Robotic Operating System (ROS) simulation and development environment to all students independently of their laptop architecture and OS; 2) provide a seamless "simulation to real" experience preserving the exciting experience of writing software interacting with the physical world; and 3) sharing GPUs across multiple student groups, thus using resources efficiently. We describe our requirements, solution design, experience working with the solution in the educational context and areas where it can be further improved. This may be of interest to other educators who may want to replicate our experience.
Soft robotic manipulators are inherently compliant thus they are ideally suited for minimally invasive diagnosis and intervention. In addition, soft robotics allows for affordable manufacturing, thus it could be adopted in low and middle-income countries where conventional robotics is prohibitively expensive. In this work, the design, manufacturing, and actuation strategy of an affordable soft robotic manipulator is presented. The manufacturing process does not rely on sophisticated technologies, and the pneumatic actuation does not require digital pressure regulators. Instead, a low-cost solution consisting of a needle valve operated by a servo motor is employed. The prototype is assessed with experiments that demonstrate its functionality.
This paper introduces the current development of distributed machine learning, federated learning and big data platform, then proposes a virtual subsystem architecture of intelligent data platform based on algorithm model, data model, resource model and security model. At the same time, the technical details of resource model scheduling, the definition of security model, the construction principle of virtual subsystem are stated. In addition, this paper also introduces the practical application of the architecture from two aspects: subsystem construction and data processing flow.
Noise in low-light image enhancement seriously affects human observation and computer vision algorithms, while existing methods inevitably introduce under- or over-contrast problems. We proposed an adaptive low-light image enhancement with decomposition denoising to perform a good visual effect. Firstly, the low-rank denoising algorithm is designed to decrease the noise of an input low-light image. And then, adaptive luminance enhancement is developed by generating a nonlinear function to improve the luminance of the denoise image. Finally, a color restoration method based on the relationship between spectral bands and input image into a color image. Subjective and objective experiments show that compared to several state-of-the-art methods, the proposed method can reasonably enhance image luminance and contrast while removing heavy noise.
UAVs rely on mapping the surroundings to gather real-time environmental information. The mapping outcome stands as the key prerequisite for further motion planning. Plenty of work has been done with sophisticated mapping algorithms. However, during UAV navigation, these algorithms are expected to be updated continuously, which consumes a significant amount of memory and computational resources in a large area. To address this limitation, in this paper, we propose an end-to-end method for UAV autonomous motion planning via Reinforcement Learning (RL). In particular, a deep RL network is built as the brain of the intelligent agent, which takes the depth image of the UAV visual feedback as input, and outputs the continuous action as a control decision. A convolutional neural network is employed to process the depth image. In order to implement and validate the proposed method, a high-fidelity 3D simulation environment is established in AirSim, which generates the real-time flight status and depth images during the UAV flight. As a result, the flight simulation demonstrates the effectiveness and efficiency of the RL-based motion planning algorithm in a complex environment. Importantly, the agent trained by the proposed IDDPG could get closer to the destination than that trained by DDPG by about 17 meters on average. Last, the computation time for each step is significantly reduced to 5 ms compared to the classical approach.
Aiming at the problem that the spatial features of pedestrian images are not aligned in current pedestrian re-identification and the network model cannot fully express the pedestrian information due to pose changes and occlusions, a method based on spatial transformation and multi-methods feature fusion is proposed. Firstly, for the pedestrian re-identification system, a processing method for enhancing the retrieval of pedestrians is provided, and the pedestrian images with more noise to be retrieved are denoised by means of side-window filtering; secondly, the spatial transformation network is improved. Channel attention and self-constrained branches are introduced to automatically align pedestrian spatial features to solve the problem of inconsistency in spatial semantic information caused by unaligned pedestrian image regions; then, multi-scale features are extracted from different deep layers of the backbone network, and coordinates attention and batch normalization of instances are integrated into different deep branches. Finally, the features of each branch are fused to obtain feature information with high representation ability. During the network training process, the dual loss function strategy is used to jointly optimize the model. Multiple experiments show that the proposed method has a higher recognition rate than other existing methods.
The use of power line communication (PLC) within a large-scale battery will allow for smart cells to communicate within a decentralised system, with an external battery management system (BMS), and also with an external smart grid network. By using PLC, the smart battery is further enhanced by allowing the BMS real-time access to in-situ cell sensor data, without the need of an additional wire harness within the battery. This paper presents experimental studies of a PLC system on four distinct lithium-ion battery pack configurations, in order to determine its suitability and limitations for large-scale energy storage systems such as for use in smart grids, battery electric vehicles, and robotic systems. Quadrature amplitude modulation (QAM) is tested up to 1024-QAM for its benefits in high bit rate communication. Recommendations on the parameters of this PLC system based upon experimental results are presented.
Swarm robots, which are inspired from the way insects behave collectively in order to achieve a common goal, have become a major part of research with applications involving search and rescue, area exploration, surveillance etc. In this paper, we present a swarm of robots that do not require individual extrinsic sensors to sense the environment but instead use a single central camera to locate and map the swarm. The robots can be easily built using readily available components with the main chassis being 3D printed, making the system low-cost, low-maintenance, and easy to replicate. We describe Zutu's hardware and software architecture, the algorithms to map the robots to the real world, and some experiments conducted using four of our robots. Eventually, we conclude the possible applications of our system in research, education, and industries.
A context can be used as an embedding extracted from historical trajectories of dynamic systems to provide meaningful information for reinforcement learning (RL) agents, thus improving the domain adaptability and robustness of RL method. However, the process of context extraction involves two key issues: How to efficiently train an encoder to extract context information from historical trajectories? And how to ensure that context information can distinguish different dynamics clearly? To tackle the problems above, a dynamics-aware context representation reinforcement learning (DacRL) is proposed in this study. We leverage the Cycle-Consistent VAE method to extract a meaningful context from historical trajectories and then divide it into domain-specific and domain-general embedding. Furthermore, we consider the contrastive nature between different tasks and use it to improve the quality of domain-specific information, so that it can represent dynamics more clearly. Finally, the current state combined with the domain-specific information is delivered into the RL agent, so as to improve the generalization of the RL agent. The simulation results illustrate that the proposed DacRL is superior to other baselines.
Matrix completion is a technique that utilizes the low rank characteristic of a matrix to recover unknown elements of the matrix with some observed elements. In practical applications, the observed elements are often contaminated by noise, which makes the matrix completion more difficult. In this paper, we first establish the mathematical models of the matrix completion with noise, and investigate three classical methods including SVT, ADM and OptSpace. Particularly, we propose to add convergence criterion and restrain divergence criterion to strengthen the constraint of the stopping criterion, solving the problem of non-convergence caused by noise data when the relative recovery error is used as the stopping criterion. Several experiments are designed to estimate the performance of these three algorithms in the task of matrix completion with noise on the synthesized dataset and Jester Joke dataset. Experimental results show that it is risky to arbitrarily assume that one of these methods is optimal. Specifically, SVT has the advantage in cost, OptSpace has the advantage in anti-noise, and ADM has the better robustness to rank. Furthermore, we apply ADM algorithm in the image recovery task, and obtain excellent results.
The integrated navigation of strapdown inertial navigation system (SINS)/ ultra-short baseline (USBL) has been widely used for the high precision navigation and positioning technology of Autonomous Underwater Vehicle (AUV). The tightly coupled method that directly uses the original information measured by USBL has high accuracy, which reduces the error in the process of position calculation in loosely coupled method. However, affected by the complex underwater acoustic environment, the statistical characteristics of USBL measurement are unknown or time-varying, which also contain irregular outliers. To solve the above problems, a robust Kalman filter based on Huber M and variational Bayes (VB) are introduced into the tightly coupled method. To improve the accuracy of statistical characteristics with outliers, the chi-square test is introduced and the simulation results verify the effectiveness of the robust filter and the simulation results verify the effectiveness of the introduced algorithm.
Agriculture is major part of GDP (Gross Domestic Product) for most of the countries in the world. Drone based farming is trendy across different countries, it will not only help to monitor the crop field but also predict the health hazards from the plant. It will help us to plan the crop production according to storage or order requests from outsiders. IOT device installation in agriculture field are projected to experience a compound annual growth rate of 20%. As per Machina Research report published on Jan 2016, the number of connected agricultural devices is expected to grow from 13 million at the end of 2014 to 225 million by 2024. Drones are aerial vehicle applied to agro farming in order to help increase crop production and monitor crop growth. This paper describes about how drone sensing & blockchain helps a farmer, merchant, agent and farm monitor team.
The accuracy in depth estimation of the transponder calibration algorithms based on Long Baseline (LBL) technology needs to be improved, especially in the case of no Depth Gauge (DG), poor depth estimation will directly affect the spatial positioning accuracy. In order to comprehensively improve the solution accuracy of transponder calibration, this paper proposes a calibration algorithm based on Local Area Segmentation (LAS) method. According to the simulation results, LAS performs the best in depth estimation, horizontal positioning solution and spatial positioning accuracy.
This paper addresses the stabilization problem of time-varying delay systems subject to actuator saturation. A novel class of Lyapunov functional with a nonlinear metric is developed to guarantee that there exists a Lyapunov-Razumikhin function that strictly decreases for the controlled system. Such a nonlinear metric is parameterized by a multi-dimensional Taylor network with concise topology, which brings into the high-order dynamic information and guarantees real-time performance. A sum-of-squares programming approach based on the designed Lyapunov functional is formulated to deduce the stability criteria, which maximizes the estimate of the region of attraction and ensures the uniform asymptotic stability of the closed-loop system theoretically. A numerical example demonstrates the effectiveness of the proposed stabilization control scheme.
Consumer IOT is one of the fastest growing market in ICT (Information and Communication Technology) space which needs constant attention on device protection to safeguard the consumers data from cyber security attacks and threat. We elucidated the home IOT security management, framework and different topologies are required to deploy at home. The AI based models are required to predict the Malware (Botnet) attacks with limited consumer IOT resources and this AI based model could deployed along with home automation software or hardware or separately using microcontroller like Raspberry Pi. The local government IOT regulation and privacy act are playing vital role on data collection and data sharing between home IOT and cloud server.
Recently, multi- sensors fusion has achieved significant progress in the field of automobility to improve navigation and position performance. As the prerequisite of the fusion algorithm, the demand for the extrinsic calibration of multisensors is growing. To calculate the extrinsic parameter, many researches have been dedicated to the two-step method, which integrates the respective calibration in pairs. It is inefficient and incompact because of losing sight of the constrain of all sensors. With regard to remove this burden, an optimization-based IMU/Lidar/Camera co-calibration method is proposed in the paper. Firstly, the IMU/camera and IMU/lidar online calibrations are conducted, respectively. Then, the corner and surface feature points in the chessboard are associated with the coarse result and the camera/lidar constraint is constructed. Finally, construct the co-calibration optimization to refine all extrinsic parameters. We evaluate the performance of the proposed scheme in simulation and the result demonstrates that our proposed method outperforms the two-step method.
