Autonomous driving systems (ADSs) must be sufficiently tested to ensure their safety. Though various ADS testing methods have shown promising results, they are limited to a fixed vehicle characteristics setting (VCS). The impact of variations in vehicle characteristics (e.g., mass, tire friction) on the safety of ADSs has not been sufficiently and systematically studied. Such variations are often due to wear and tear, production errors and so on, which may lead to unexpected driving behaviours of ADSs. To this end, in this article, we propose a method, named SafeVar , to systematically find minimum variations to the original vehicle characteristics setting, which affect the safety of the ADS deployed on the vehicle. To evaluate the effectiveness of SafeVar , we employed two ADSs and conducted experiments with two driving scenarios. Results show that SafeVar , equipped with NSGA-II, generates more critical settings that put the vehicle into unsafe situations, as compared with the baseline algorithm. We also identified critical vehicle characteristics and reported to which extent varying their settings put the ADS vehicle into unsafe situations.

Context: Deep learning has been widely used in Autonomous Driving Systems (ADS). Though significant progress has been made regarding their efficiency and accuracy, uncertainty remains a critical factor affecting ADS safety . Such uncertainties are often due to environmental noise and/or imperfect algorithm structures. Studies on uncertainty quantification mostly focus on single classification tasks and overlook how uncertainties propagate from the perception to downstream decision-making, studying of which is critical, as the interplay between perception and decision-making can significantly impact the overall safety of ADS. Objectives: We quantify and understand the uncertainty propagation from sensor data to deep learning models, as well as its impact on ADS safety. Methods: We present an empirical study that quantifies both aleatoric and epistemic uncertainties and assesses how such uncertainties propagate and impact ADS safety under various sensor noise conditions. We also investigate the suitability of two epistemic uncertainty quantification methods (i.e., MC Dropout and Deep Ensembles) to ADS tasks and their cost-effectiveness in selecting highly-uncertain samples. Results: Results show that increased noise can significantly increase uncertainty and degrade model performance, thereby compromising decision-making and potentially impacting ADS safety. Both MC Dropout and Deep Ensembles effectively measure the model’s epistemic uncertainty, with MC Dropout showing higher correlation with ADS safety, and saving time and computational costs. Moreover, there are significant differences in the highly-uncertain samples they identified. Conclusion: Our results show the importance of considering uncertainty propagation to ensure the ADS safety. Compared to Deep Ensembles, MC Dropout’s efficiency makes it a more suitable choice in the context of ADS.

Robotic micromanipulation systems (RMS) enable precise and repeatable operations under a microscope. Traditional RMS rely solely on microscopic visual feedback, necessitating time-consuming manual positioning to bring the tool tip within the microscope field-of-view (Micro-FOV), which limits efficiency and heavily depends on operator skill. This paper proposes an innovative RMS that integrates macro and micro vision to automate the aforementioned tool tip positioning and facilitate resilient control. The system utilizes an external camera to obtain the macro field-of-view (Macro-FOV), containing the tool and fiducial markers, and estimates the tool tip’s 3D position by triangulation. Visual servoing is then used to guide the tool tip towards the Micro-FOV. Under the Micro-FOV, a tool-sweep detector based on partitioned difference images is used to sequentially locate the tool’s shaft and tip. After auto-focusing, the system executes tool tip and Petri dish resilient control based on our developed self-calibration and self-recalibration mechanisms. During the operation, the system provides an intuitive user interface that includes both macro and micro information, improving the visualization and productivity of micromanipulation. Experiments show that the self-initialization scheme can be implemented across different macro camera viewpoints, reducing the average tip positioning time from 65.70 to 50.08 seconds compared to manual operation, thereby decreasing manual labor intensity and improving efficiency. The self-recalibration mechanism achieves precision and resilient control, with an average error of$0.95~\mu $m over 25 continuous trials. Additionally, the system exhibits robustness against vibration and visual interference, underscoring its potential for diverse biomedical applications.Note to Practitioners—In the biomedical field, robotic micromanipulation systems (RMS) are valued for their high precision and repeatability. Existing research on RMS typically focuses on achieving varying degrees of automation using visual feedback from the microscope, with the assumption that the tool tip is already within the Micro-FOV prior to the task. However, the initial step of moving the tool tip from the Macro-FOV to the Micro-FOV and eventually bringing it to focus usually requires time-consuming manual operation that highly depends on the skill of individual operators. To achieve automatic positioning and resilient control of the tool tip, this paper presents an RMS that integrates macro and micro vision. The system combines visual feedback from a macro camera and a microscope to complete the self-initialization of the system in three steps: macro tip positioning, micro tip positioning, and self-calibration. After self-initialization, the system provides the operator with an intuitive cursor-based user interface. During operation, the system uses visual feedback to detect the tip position and ensures the control accuracy through a self-recalibration mechanism. The proposed self-initialization method has the potential for a wide range of applications. It can be extended to micromanipulation systems with different types of end-effectors, thereby improving the efficiency and precision of micromanipulation.

Accurate classification of encrypted traffic is vital for effective network management and Quality of Service (QoS). Protocol identification and application recognition are central to this process. Existing solutions generally use separate models for protocol identification and application recognition. However, running these models in parallel in online environments significantly increases system complexity, resource consumption, and maintenance costs. Additionally, these models often suffer from poor traffic representation by discarding crucial byte information and retaining irrelevant biases. This paper introduces an efficient multi-task learning framework, ELEMTC, designed to simultaneously perform protocol identification and application recognition, while offering a tailored traffic representation scheme for network management. To improve efficiency, we propose a novel sample pre-training task, Detection Token Replacement, specifically tailored for the networking domain. Experimental results on the ISCX VPN-non VPN dataset show that ELEMTC achieves Fl scores of 98.37% and 99.46% for application recognition and protocol identification, respectively. These results demonstrate that ELEMTC significantly outperforms current state-of-the-art methods. Furthermore, ELEMTC demonstrates outstanding efficiency by improving inference speed 20 times while maintaining relatively low memory usage.

Summary Collaborative and autonomous driving vehicles combine hardware and software complex processes, also are heavily dependent on and influenced by the world of physical and cyber interactions. They have enabled many new features and advanced functionalities, such as stochastic and hybrid natures, mobile spatial topologies, and time‐critical dependability. However, the existing modeling and verification techniques have not established faith in proving correctness and safety. Spatial and time collision avoidance remains crucial obstacles on the path to becoming ubiquitous and dependable. In order to ensure safety, we first design an accident prediction architecture in system design‐time and run‐time stages. We apply it on collaborative and autonomous overtaking systems involving spatial‐ and time‐critical accident predictions. Then, we develop a novel and dedicated spatio‐clock stochastic specification language (SCSSL) to describe safety invariants and guards in domain‐specific autonomous driving systems. Next, we create the spatio‐clock stochastic and hybrid automata models based on SCSSL in order to model inherently stochastic and hybrid behaviors. To illustrate the effectiveness of spatio‐clock consistency stochastic specification and verification, we adopt statistical model checking natively to provide reliable predictions for the incoming collision instants and positions. Finally, we present an illustrative overtaking case study to verify spatio‐clock stochastic and hybrid related properties and ensure correct modeling, and demonstrate the significance of our proposed approach.

Autonomous vehicles must operate safely in their dynamic and continuously-changing environment. However, the operating environment of an autonomous vehicle is complicated and full of various types of uncertainties. Additionally, the operating environment has many configurations, including static and dynamic obstacles with which an autonomous vehicle must avoid collisions. Though various approaches targeting environment configuration for autonomous vehicles have shown promising results, their effectiveness in dealing with a continuous-changing environment is limited. Thus, it is essential to learn realistic environment configurations of continuously-changing environment, under which an autonomous vehicle should be tested regarding its ability to avoid collisions. Featured with agents dynamically interacting with the environment, Reinforcement Learning (RL) has shown great potential in dealing with complicated problems requiring adapting to the environment. To this end, we present an RL-based environment configuration learning approach, i.e.,DeepCollision, which intelligently learns environment configurations that lead an autonomous vehicle to crash. DeepCollision employs Deep Q-Learning as the RL solution, and selectscollision probabilityas the safety measure, to construct the reward function. We trained four DeepCollision models and conducted an experiment to compare them with two baselines, i.e., random and greedy. Results show that DeepCollision demonstrated significantly better effectiveness in generating collisions compared with the baselines. We also provide recommendations on configuring DeepCollision with the most suitable time interval based on different road structures.

Inventory Routing Problem (IRP) is a typical optimization problem in logistics. To reduce the total cost, which contains the product transportation cost, the inventory holding cost, the customer satisfaction cost, etc., a wide range of impact factors have to be taken into consideration. Since more and more intelligent devices have been adopted in the management of modern logistics, the amount of the collected data (relevant to those impact factors) increases exponentially. However, the quality of the collected data is suffering from a certain number of uncertainties, such as device status and the transmission network environment. Considering the volume and quality of the collected data, the traditional data‐driven distribution optimization methods encounter a bottleneck. In this paper, we propose a hybrid optimization method which combines data‐driven and knowledge‐driven techniques together. In our method, a domain ontology, which has better scalability and generality, is built as an extension of data‐driven optimization algorithms. Knowledge reasoning techniques are also combined to handle data quality issue and uncertainties. To evaluate the performance of our method, we carried out a case study, which is provided by a French company “Pierre Fabre Dermo‐Cosmetics” (PFDC). This case study is a simplified scenario of the practical business process of PFDC.

Cyber-physical systems (CPSs) have emerged as a potential enabling technology to handle the challenges in social and economic sustainable development. Since it was proposed in 2006, intensive research has been conducted, showing that the construction of a CPS is a hard and complex engineering process due to the nature of integrating a large number of heterogeneous subsystems. Among other approaches to dealing with the complex design issues, model-driven design of CPSs has shown its advantages. In this review paper, we present a survey of research on model-driven development of CPSs. We are concerned mainly with the widely used methods, techniques, and tools, and discuss how these are applied to CPSs. We also present comparative analyses on the surveyed techniques and tools from various perspectives, including their modeling languages, functionalities, and the challenges which they address in CPS design. With our understanding of the surveyed methods, we believe that model-driven approaches are an inevitable choice in building CPSs and further research effort is needed in the development of model-driven theories, techniques, and tools. We also argue that a unified modeling platform is needed. Such a platform would benefit research in the academic community and practical development in industry, and improve the collaboration between these two communities.

Multi-organization collaboration, which allows partners focus on their core business, is becoming a trend. Besides the interoperability of each partner, the mechanism of selecting qualified and suitable partners is another key issue to guarantee the success of collaboration. Based on our previous work, this paper aims to provide a model-driven approach to solve the partners selecting problem in building collaboration. In this approach, a meta-model is defined to describe the context of collaboration. All the potential partners’ inputs shall be conformed to this meta-model. In order to select the required partners automatically, semantic checks are combined.

International audience

With the rapid advance of mobile computing technology and wireless networking, there is a significant increase of mobile subscriptions.This brings new business requirements and demands in mobile software testing, and causes new issues and challenges in mobile testing and automation.As there are multiple platforms for diverse devices, engineers suffer from the different scripting languages to write platform-specific test scripts.In addition, a unified automation infrastructure is not offered with the existing test platform.This paper proposes a novel approach to mobile application testing based on natural language scripting.A Java-based test script generation approach is developed to support executable test script generation based on the given natural language-based mobile app test operation scripts.A prototype tool is implemented based on some open sources.Finally, the paper reports empirical studies to indicate the feasibility and effectiveness of the proposed approach.

Face recognition system is a widely-used intelligent application nowadays.Existing recognition system evaluation methods primarily focus on recognition rate, i.e., the correct result.However, current research seldom focuses on the quality evaluation of face recognition systems.They seldom consider accuracy or the quality of recognition.To address this issue, this paper proposes several quality factors for evaluation.In addition, corresponding metrics for diverse quality factors are illustrated.Moreover, the paper presents an experimental study on a realistic non-trial face age recognition system using the proposed quality evaluation method.The study result shows the proposed method is feasible and effectiveness in quality evaluation.

Several standards have been defined for enhancing the efficiency of B2B web-supported collaboration. However, they suffer from the lack of a general semantic representation, which leaves aside the promise of deducing automatically the inter-enterprise business processes. To achieve the automatic deduction, this paper presents a social platform, which aims at acquiring knowledge from users and linking the acquired knowledge with the one maintained on the platform. Based on this linkage, this platform aims at deducing automatically cross-organizational business processes (i.e. selection of partners and sequencing of their activities) to fulfill any opportunity of collaboration.

Web service composition, as one of the key aspects in web service domain, has attracted more and more research attentions. Generally, in order to provide a powerful function to a specific problematic, several web services should combine and work together. Such a collaboration of web services is regarded as web service composition. There are two main difficulties in web service composition: selecting web services as partners and making interactions among these web services. A web service works as a functional black box, it takes in inputs and generates outputs. For a specific web services, both the inputs and the outputs are in specific formats. In order to make interactions among web services, it is necessary to be synergistic among their inputs and outputs. To generate specific inputs for a particular web service, the outputs from one or several other web services should transform the formats and combine together. This paper presents an automatic model transformation methodology, which focuses on transforming and combining outputs to generate inputs for web services. This automatic model transformation methodology regards all web services' inputs and outputs as models. In order to do the transformation and combination process efficiently and effectively, syntactic checking and semantic checking measurements have been combined into a refined model transformation process.

Modelling and model transformation are regarded as two pillars of model-driven engineering; they have been used together to solve practical problems. For instance, since different models (e.g. data model) are used by heterogeneous partners involved in a specific collaborative situation, there is an urgent need for model transformations to exchange information among the heterogeneous partners. To quickly define model transformations, this paper presents an approach, which could replace the users' effort in making mappings during the definition of a model transformation process. This approach is based on model transformation methodology, using syntax and semantic relationship among model elements. For this, a generic meta-metamodel and semantics checking methodology are proposed, before being illustrated by an example.