Urban mobility is undergoing a profound transformation driven by multiple forces, including sustainability, autonomous vehicles, and safety. In this study, we focus on road safety, which remains a global concern with 1.19 million road traffic deaths reported annually by the World Health Organization. Among the key factors contributing to crash risk, infrastructure design has been recognized as one of the most actionable. This paper proposes an AI-based tool that provides automatic urban risk assessment through topological imagery, even in the absence of detailed crash records in a specific area of interest. We develop a regression model based on a modified ResNet-18 architecture, trained to predict a risk index from map-based images of urban areas and real-world telematics data, specifically harsh braking events. The model is trained on a dataset of over 80,000 harsh events collected from approximately 10,000 vehicles in the city of Milan during 2024.The results show that the model can effectively infer infrastructure-related road risk from topological inputs only, providing a lightweight and data-efficient prior for intelligent vehicles, and supporting enhanced global path planning in an autonomous driving context, even in scenarios with limited or no incident data.

Accurately localizing 3D objects like pedestrians, cyclists, and other vehicles is essential in Autonomous Driving. To ensure high detection performance, Autonomous Vehicles complement RGB cameras with LiDAR sensors, but effectively combining these data sources for 3D object detection remains challenging. We propose LCF3D, a novel sensor fusion framework that combines a 2D object detector on RGB images with a 3D object detector on LiDAR point clouds. By leveraging multimodal fusion principles, we compensate for inaccuracies in the LiDAR object detection network. Our solution combines two key principles: (i) late fusion , to reduce LiDAR False Positives by matching LiDAR 3D detections with RGB 2D detections and filtering out unmatched LiDAR detections; and (ii) cascade fusion , to recover missed objects from LiDAR by generating new 3D frustum proposals corresponding to unmatched RGB detections. Experiments show that LCF3D is beneficial for domain generalization, as it turns out to be successful in handling different sensor configurations between training and testing domains. LCF3D achieves significant improvements over LiDAR-based methods, particularly for challenging categories like pedestrians and cyclists in the KITTI dataset, as well as motorcycles and bicycles in nuScenes. Code can be downloaded from: https://github.com/CarloSgaravatti/LCF3D .

Connected and Automated Vehicles (CAVs) are revolutionizing road transport by offering enhanced safety, efficiency, and sustainability. A key requirement for their safe operation on roads is the availability of highly accurate positioning information with ultra-low latency. This study presents the design and assessment of an infrastructure-based vehicle positioning system, focusing on the latency involved in transmitting position information to vehicles using Vehicle-to-Everything (V2X) connectivity. Specifically, we consider a roadside positioning infrastructure that integrates an Ultra Wideband (UWB) technology for positioning and an ITS-G5 V2X connectivity for communication. We measure and analyze the round trip time of the V2X communication link to gain insights on the latency performance. The positioning performance is also analyzed by comparing the trajectory followed by the vehicle with the one planned by the onboard control system. On-field evaluations are conducted in a highway tunnel, demonstrating the ability of the infrastructure localization system in successfully enabling autonomous navigation.

The Usage-Based Insurance paradigm, which is receiving a lot of attention in recent years, envisages computing the car policy premium based on accident risk probability, evaluated observing the past driving history and habits. However, Usage-Based Insurance strategies are usually based on simple empirical decision rules built on travelled distance. The development of intelligent systems for smart risk prediction using the stored overall driving behaviour, without the need of other insurance or socio-demographic information, is still an open challenge. This work aims at exploring a comprehensive machine learning-based approach solely based on driving-related data of private vehicles. The anonymized dataset employed in this study is provided by the telematics company UnipolTech, and contains space/time densely measured data related to trips of almost 100000 vehicles uniformly spread on the Italian territory, recorded every 2 km by on-board telematics fix devices (black boxes), from February 2018 to February 2020. An innovative feature engineering process is proposed, with the aim of uncovering novel informative quantities able to disclose complex aspects of driving behaviour. Recent and powerful learning techniques are explored to develop advanced predictive models, able to provide a reliable accident probability for each vehicle, automatically managing the critical imbalance intrinsically peculiar this kind of datasets.

The paper describes the design and validation of a spatial image registration algorithm for a vehicular Head-Mounted Augmented Reality (AR) system. AR can considerably improve the driving experience by increasing the driver’s situational awareness. AR can only work if stable and realistic holograms are generated. The process of generating the holograms so that they appear in a specific position in the world is also known as image registration. Since AR devices employ see-through Head-Mounted Displays, realistic image registration requires high-accuracy head tracking. Solutions exist in static environments where state-of-the-art simultaneous localization and mapping algorithms suffice. Vehicles are more challenging. In aerospace, costly optical-inertial tracking systems are regularly employed. This paper focuses instead on low-cost ground vehicles and proposes a solution that does not require aerospace-grade Inertial Measurement Units and is easily integrable on cars. The proposed solution, tested on a racing circuit, is based on passive markers and on the stereoscopic detection of the road plane on which the AR features are anchored.

Powered two-wheelers exhibit complex dynamics, primarily due to their broader range of possible attitude configurations compared to four-wheelers. This is especially true when considering accident scenarios, in which the dynamics can be affected by both high-frequency events, such as in high-sides, marked by rapid variations of the characteristics signals, and low-frequency events, such as in side slips, which exhibit slower and monotonic change in the same quantities. Therefore, designing an effective automatic eCall system for this type of vehicle is particularly challenging, while at the same time, it can significantly increase safety in dangerous situations. This paper proposed a novel approach for detecting falls of two-wheeled vehicles that avoids missed detections while minimizing the number of false alarms, which would heavily undermine its usability. We show, based on extensive analysis of a wide range of experimental data, why and how the proposed approach allows us to overcome the limitations of the existing proposal, ensuring a consistent detection of the wide range of falls that can happen on two-wheelers, also providing a detailed comparison with the scientific literature.

In this article, we propose a path planning algorithm focused on high-speed lane-change scenarios in autonomous racing applications. Specifically, by formulating a quadratic Optimal Control Problem, we design nearly curvature optimal trajectories that can allow for high-speed maneuver execution. The problem formulation includes a convenient approach for the linearization and discretization of the path system dynamics. Unlike most state-of-the-art approaches, the proposed method does not need to constrain the path shape to pre-defined polynomial primitives and does not perform inefficient grid searches over sets of candidate maneuvers. Instead, it can directly optimize path curvature, while fulfilling path smoothness and trajectory constraints. We tested the proposed methods in a racing simulation environment, and compared its results with a benchmark algorithm. The results show the algorithm’s ability to return nearly curvature-optimal solutions while maintaining real-time feasibility. Eventually, such results demonstrate the effectiveness of the proposed solutions.

This paper presents a new approach for automatic vehicle model recognition and simultaneous estimation of lateral transit position, based on magnetic sensor technology. A set of magnetic sensors is deployed on the road surface and, upon transit of a target vehicle on the equipment, the system records six magnetic signatures relative to different vehicle sections. The recorded signatures are then compared with the Dynamic Time Warping algorithm to previously recorded ones, which are relative to known vehicles that have transited at known lateral position; the system then assesses whether the target vehicle's model matches one of the models already in the database, and estimates its lateral transit position. With the considered experimental set-up, the system is able to discriminate between many different vehicle models and six lateral positions, with a resolution of about 20 cm: the performance of the system is presented by comparing a set of different classifiers. In terms of vehicle model recognition, 1-Nearest Neighbor classifier obtains 0% of misclassification rate, while for lateral position estimation, if an error of one position is tolerated (precision of ± 20 cm), the system is shown to reach 2.4% of misclassification rate.

Assessing the driving-style from dynamic data is a well established line of research, which has tackled the description of risky behaviours, the profiling of energy-consumption habits, and the detection of different driver's characteristics from the analysis of motion data. In the last years, as smartphone ownership has become widespread, such an assessment has been increasingly relying on the measurements taken from the inertial sensors on board of the smartphone itself. This work stands in this context, and it aims to design a 4-dimensional driving-style assessment for insurance purposes. The main contribution is adding, to more common proxies of risky-driving, the dimension of smartphone usage, the detection of which is performed through an appropriate processing of smartphone-based inertial sensors, thus not relying on privacy-sensitive monitoring of phone usage behavior. Physics-based, fine-grained dynamic features are used to classify the overall riskiness of the driving-style, thus providing a comprehensive insight into the most discriminating features. The study is based on experimental data, collected over more than five thousands kilometers of varied car trips.

Connected and automated vehicles (CAVs) have the potential to improve the safety of automated driving by utilizing increased awareness about their surroundings in real time vehicle control. In this paper we propose a framework for a shared perception system suitable for CAVs and explain the algorithms used in the system. Finally, we experimentally demonstrate the benefit of our shared perception system for automated vehicles in uncertain environments.

On vehicles there exists a close coupling between brake and suspension dynamics, making semi-active damper control a promising way for brake maneuver optimization, which has been widely researched in the automotive and motorcycle field. Experimental data of bicycle dynamics analyzed in this paper show substantial differences to classical vehicle dynamics. By first-principle control-oriented modeling and full vehicle nonlinear multibody simulation it is shown that this can be traced back to two phenomena: the dynamic rider response and fork bending. These are very general effects, but crucial for vehicle dynamics control on bicycles, one of the most widely used means of transportation.

Suspension systems on bicycles have a tendency to severe brake-induced dive-in, caused by the small wheelbase in combination with a high center of gravity. Semi-active dampers allow the implementation of anti-dive functionality, preventing this behavior. Experimental analysis has shown that this yields significant advantages during brake control on level surfaces. In the presence of additional road excitation, however, a strong conflict arises. A specific test case is a bump occurring while braking, when the damping is set to the hardest value in order to mitigate dive-in. A simulative analysis illustrates that especially the dynamic wheel load is affected, which during braking is safety critical. By simulation and experimental implementation it is shown that using a simple semi-active control rule a decent trade-off can be found. Finally, the influence of the actuator response time is evaluated.

The main objective of this paper is to determine the feasibility of designing a driver drunkenness detection system based on the dynamic analysis of a subject's pupillary light reflex (PLR). This involuntary reaction is widely utilized in the medical field to diagnose a variety of diseases, and in this paper, the effectiveness of such a method to reveal an impairment condition due to alcohol abuse is evaluated. The test method consists in applying a light stimulus to one eye of the subject and to capture the dynamics of constriction of both eyes; for extracting the pupil size profiles from the video sequences, a two-step methodology is described, where in the first phase, the iris/pupil search within the image is performed, and in the second stage, the image is cropped to perform pupil detection on a smaller image to improve time efficiency. The undesired pupil dynamics arising in the PLR are defined and evaluated; a spontaneous oscillation of the pupil diameter is observed in the range [0, 2] Hz and the accommodation reflex causes pupil constriction of about 10% of the iris diameter. A database of pupillary light responses is acquired on different subjects in baseline condition and after alcohol consumption, and for each one, a first-order model is identified. A set of features is introduced to compare the two populations of responses and is used to design a support vector machine classifier to discriminate between “Sober” and “Drunk” states.

Despite the enormous progress of the last years, urban environments still represent a challenge for robot autonomous navigation. This paper focuses on the problem of detecting street pole-like obstacles using a monocular camera. Such obstacles, due to their thin structure, may be difficult to be detected by common active sensors like lasers. This is even more critical for innovative solid state LiDARs like the one employed in this work because, at the actual state, they are characterized by very low angular resolutions. The approach described here is based on identifying poles as long vertical structures in the image and in locating them with respect to the robot using a Kalman filter based depth estimation. This information can then be fused with the information coming from LiDARs realizing a complete obstacle detection module.

Smartphone usage while driving is unanimously considered to be a really dangerous habit due to strong correlation with road accidents. In this paper, the problem of detecting whether the driver is using the phone during a trip is addressed. To do this, high-frequency data from the triaxial inertial measurement unit (IMU) integrated in almost all modern phone is processed without relying on external inputs so as to provide a self-contained approach. By resorting to a frequency-domain analysis, it is possible to extract from the raw signals the useful information needed to detect when the driver is using the phone, without being affected by the effects that vehicle motion has on the same signals. The selected features are used to train a Support Vector Machine (SVM) algorithm. The performance of the proposed approach are analyzed and tested on experimental data collected during mixed naturalistic driving scenarios, proving the effectiveness of the proposed approach.

This paper addresses the problem of the development, implementation and testing of a Lane Change Decision Aid System on a motorcycle, by using a short range radar sensor and a set of LEDs to interface with the driver. First, a feasibility analysis for such application is performed, then the algorithm is described, and finally the results of on-road tests are presented to illustrate and validate the method. The algorithm is composed by a first block for stabilizing the detected objects through a finite state machine and filtering the data coming from the sensor by implementing a Kalman filter that reduces the sensor's inaccuracies. A decision block sets the state of the system by checking the position and speed of the detected objects, and finally an HMI block computes the Hazard Level considering position and Time-to-Collision of the detected objects and delivers a warning to the driver.

This paper presents a novel haptic support system for braking control in road bicycles. The basic idea is to provide a direct indication of the wheel deceleration to the rider in form of a brake lever vibration. The more intense the vibration, the closer the rider is to the threshold safe deceleration. This enables an efficient information transfer and an easy, safe deceleration. The paper presents the rationale and the proposed system. The performance is assessed through a number of experimental tests, showing that the presence of the haptic feedback enables the rider to brake more safely and more consistently than without haptic feedback.

Electric vehicles (EVs) are recognized to be an effective way to reduce global warming pollution. Therefore, there is an increasing interest and effort in developing new ideas and technology to spread the use of EVs as an alternative to conventional cars. However, the consumers' perception of EVs, especially concerning their limited and not-fully predictable range, is still of low reliability, thus the number of sales is increasing slower than expected. In this paper, a software tool for non-EVs owners is presented, with the aim of profiling the drivers and show which EV alternative, if any, would be suitable for each case. The final goal is to encourage electric mobility against range anxiety. The whole analysis is carried out by using experimental data collected from some users in Milan, Italy.

This paper deals with collision and hazard detection for motorcycles via inertial measurements. For this kind of vehicles, the most difficult challenge is to distinguish road's anomalies from real hazards. This is usually done by setting absolute thresholds on the accelerometer measurements. These thresholds are heuristically tuned from expensive crash tests. This empirical method is expensive and not intuitive when the number of signals to deal with grows. We propose a method based on self-organized neural networks that can deal with a large number of inputs from different types of sensors. The method uses accelerometer and gyro measurements. The proposed approach is capable of recognizing dangerous conditions although no crash test is needed for training. The method is tested in a simulation environment; the comparison with a benchmark method shows the advantages of the proposed approach.

This paper addresses the problem of motor activation on electric-power-assisted cycles that are not equipped with pedal torque sensors. The algorithm is based on inertial and motor velocity measurements along with pedal cadence. The problem is solved by a system that detects when the rider is on the saddle. Three different dynamics (longitudinal, lateral, and vertical) are analyzed, showing differences in their characteristics when the rider is on the saddle and when she is walking the bicycle. The problem is framed as a fault detection problem based on three independent analyses that are then fused. Experimental validation on a single-gear bicycle shows that the proposed solution outperforms the traditional cadence-based activation logic both reliability and reactiveness standpoints. The proposed system improves reactiveness compared with the standard cadence-based method without affecting safety.

This paper addresses the issue of data-driven online velocity optimization of an autonomous sailboat. Autonomous sailboats represent an ideal for long range and duration reconnaissance missions. Sailboat control is a challenging control task; sailboats are characterized by a number of control variables, all of which affect the ship trajectory and state in a highly nonlinear fashion. In this paper, a path-following automatic sailboat controller is presented. The control system has two main components: a heading control, acting on the rudder, and a velocity optimizer, acting on the sails. The optimizer is based on a modified extremum seeking approach. This paper also derives a first-principle-based 4 DoF sailboat model that is experimentally validated and used to guide the design and tuning of the control system. In fact, the control system is first tuned and validated in simulation. The simulation environment enables the comparison of the proposed model against a theoretical benchmark and a state-of-the-art controller. The analysis reveals that the proposed control system achieves near-optimal performance and considerably outperforms the state-of-the-art solution. Finally, the controller is tested and validated on an instrumented scale model.

Range extenders (REs) increase the driving range of electric vehicles (EVs) at the price of additional weight and encumbrance, which are unnecessary in normal urban usage. An innovative concept of an extended range EV is studied here, i.e., an EV that can exploit a rental RE only when necessary to complete a mission. The energy management system not only dispatches power between a battery and an RE but also decides whether or not and when to rent the RE. This paper investigates the optimal control policy that minimizes the monetary cost attained by the user. A simulation analysis carried out in a representative set of scenarios demonstrates the interest of this formulation and shows that mixed-integer convex programming (MI-CP) attains high accuracy in a fraction of the time required by dynamic programming, with a negligible loss of performance. In a real-time framework, the MI-CP core is fed by a forecast of the mission and integrated with a power dispatch strategy that optimizes local performance. The simulation study shows that the resulting policy approaches the global optimum.

The free-floating car sharing model is a recently introduced vehicle rental model, which allows customers to return the car anywhere within the operation area, without relying on depot stations. Driven by the flexibility of such a model, the popularity of car sharing has increased rapidly during the last years. However, some critical issues still arise when a user needs to make plans of vehicle usage, since no information is available on future vehicle locations. In this paper, the Vehicle Distance Prediction (VDP) approach is proposed, aimed to predict the distance of the nearest available vehicle at a given future instant. This technique shows great potential also for the service manager, e.g. vehicles could be moved in advance by the staff to balance the fleet distribution. The effectiveness of the proposed prediction approach is assessed on a real dataset taken from a car sharing service in Milan, Italy.

This paper presents the design of the vehicle platform and of the related control system of an innovative electric vehicle tailored to indoor personal mobility. The vehicle is suitable for the transportation of a single passenger or small loads. It has no handlebars; therefore, the rider stands in an upright position and controls the vehicle direction via the combined use of a smartphone and of weight balancing. Specifically, the inertial sensors on board the smartphone allow commanding the vehicle motion, and two metal insoles equipped with pressure sensors allow gathering the user's weight distribution in real time to issue the steering commands. This paper presents all the development phases, characterized by a codesign of the vehicle mechanics and electronic systems that makes the motion control problem more easily managed than that of comparable existing mobility solutions.

Lithium ion (Li-ion) batteries are the standard choice for many applications; but their behavior is complex. In order to safely and efficiently exploit their advantages, advanced model-based Battery Management Systems (BMS) are required. This paper introduces a computationally efficient, control-oriented model for a Li-ion cell. The model, by augmenting the classical Randle model with diffusive dynamics, is capable of describing both the high and low frequency behavior of the cell. The identification of the proposed model is detailed and an identification protocol proposed. The model is validated on a commercial lithium-ion cell. The proposed model yields an efficient simulation tool that can be employed for BMS design.

Energy-consumption in ground transportation systems is a major cause of CO2emissions and related environmental concerns. A relevant part of such an energy use can be saved by devising systems that can help promoting an economical driving style. To address this challenging issue, this paper presents a method and system that allow providing a quantitative description the driving style and illustrating it to the driver by means of real time visual feedback. The driving style assessment is based on the definition of appropriate energy-oriented cost functions that can be evaluated based on inertial measurements only, thereby providing a vehicle-independent methodology. The effectiveness of the approach, and the savings enabled by the interaction with the driver are assessed with an experimental campaign carried out on urban and extra-urban routes by different drivers.

Lean angle is a crucial variable in determining the dynamic behavior of two-wheeled vehicles. As such, its knowledge is needed to enhance active safety via traction, braking and stability control systems. To estimate the roll angle, this paper proposes a method that employs a reduced set of sensors with respect to existing solutions; namely, two accelerometers and a gyroscope are needed. Notably, it does not require the knowledge of the speed signal, which can be unavailable and/or affected by significant uncertainties when large accelerations/decelerations occur. The estimation performance are analyzed based on experimental data collected on a sport motorcycle at the Imola circuit.

This work proposes a novel method to estimate an index which summarizes the state of the satellite constellation in a Global Positioning System (GPS). Experimental results evidence that it is correlated with the localization error. An algorithm for the estimation and the correction of the GPS offset is therefore proposed. The improvement of the localization accuracy is demonstrated using an automotive, high-end GPS.

New driveline architectures equipped with torque-biasing devices such as active differentials and active transfer cases have yielded a new generation of on-demand four-wheeldrive vehicles, where the torque distribution between left and right and between front and rear axles can actively be modulated online. This allows one to design active vehicle-control systems that are capable of altering, via electronic control, the behavior of a car dictated from its mechanical layout, e.g., understeering and oversteering characteristics. This paper proposes a control strategy that optimizes vehicle performance while guaranteeing vehicle stability and drivability by actively controlling the transfer case. The performance of the overall control strategy is assessed on both a multibody simulator and an instrumented test vehicle.

In this paper the analysis and Data-Mining of a large data-set related to a very popular Italian Virtual Community is presented. The Community is constituted by more than half-million registered users, characterized by a unique nickname. Each user has its own profile, which is filled during the registration procedure, on a voluntary basis. Two data-sets are used: the Data-Base of the Users, and the log-file of the servers hosting the Community web-site. This work is constituted by three main parts: 1) analysis and clustering of the Users DB; 2) analysis and clustering of the navigation sessions; 3) correlation of Users clusters and navigation sessions clusters. This analysis provides a complete and full-rounded picture of the Virtual Community Users.

1 Introduction The problem this paper focuses on is the classical problem of unsupervised clustering of a data-set. In particular, the bisecting divisive clustering approach is here considered. This approach consists in recursively splitting a cluster into two sub-clusters, starting from the main data-set. This is one of the more basic and common problems in fields like pattern analysis, data mining, document retrieval, image segmentation, decision making, etc. ([13], [15]).

1 Introduction and problem statement The problem this paper focuses on is the unsupervised clustering of a data-set. The dataset is given by the matrix M = [x1,x2,…,xN] ∊ ℜp×N, where each column of M, xi ∊ ℜp, is a single data-point. This is one of the more basic and common problems in fields like pattern analysis, data mining, document retrieval, image segmentation, decision making, etc. ([12, 13]). The specific problem we want to solve herein is the partition of M into two sub-matrices (or sub-clusters) ML ∊ ℜp×NL and MR ∊ ℜp×NR, NL + NR = N. This problem is known as bisecting divisive clustering.