Autonomous vehicle control architecture
Human-like driving allows other participants to predict the behavior of the autonomous vehicle (AV) and provides comfort for passengers. Understanding the AV’s behavior will increase the chance of cooperation of AV and human-driven vehicles in critical situations
The architecture refers to a set of programs that, structured in a loosely coupled three-level hierarchy and connected by specific links, is mimicking the human driver behavioral model.
The Strategic level is the one that proposes an offline navigation solution. All navigation circumstances are considered to be known.
At the Tactical level the AV finds the necessary solutions to achieve the imposed task: bypass and avoid the obstacles; park; stop etc.
At the Operational level, the AV implements the obtained solutions: the action of pressing the clutch pedal followed by the action of turning the ignition key, etc.
The tactical level of the driving architecture comprises three closed-loop circuits. These circuits are highlighted in the next figure
The simulation of the tactical level was developed in MATLAB using for visualization the Unreal Engine together with the Autonomous Vehicle Toolbox
Electric vehicle simulator with driver-in-the-loop
A driving simulator with realistic interaction, operating environment, and feedback eliminates the difficulties of the road test but allows the understanding of driving behavior, testing driver assistant systems, and for traffic research.
The hardware components of the simulator are the Logitech G29 steering wheels with pedals, the Moog Motion System 6DOF 2000E motion platform, driver’s seat with a seatbelt, three high-definition monitors, and two computers (one for the dynamic model and one for the data transfer).
The movement of the motion platform is imposed with the Moog motion platform with six identical electromechanical actuators with high-performance electric motors, power supply system, and servo controllers. The platform has a base frame and a moving platform. The actuators of the Stewart platform have permanent magnet synchronous motors, belt drive, and a ball-screw mechanism, and it is controlling the position of the moving platform. The motion platform is providing six degrees of freedom (roll, pitch, heave, surge, yaw, lateral) with acceleration up to 0.6g and 500deg/sec2.
The main loop is simulating the dynamics of the electric vehicle using the input data provided by the driver. Two sets of data present special interest, on one hand, the performance parameter of the electric vehicle and on the other hand, the linear and angular accelerations of the vehicle. The first set is important in evaluating the electric vehicle (state of charge, consumption, driving range and efficiency) and the second represents the input data into the motion cueing algorithm which computes the required movement of the motion platform to reproduce with high fidelity the sensation of a moving vehicle in the simulated environment.
Interpretation of driving scene using automated salient region detection
Drivers are using visual perception and are usually focusing their attention for decision-making on the main features of the scenery ahead, like the curvature of the road ahead, neighboring vehicles, bicycles, pedestrians, and other obstacles.
The interpretation of a scene from an event-reasoning point of view using automated salient region detection is an important step in image labeling for autonomous vehicles’ behavior training and improvement.
Blockchain for automotive industry
Blockchain proves to be a powerful technology that can drive forward the automotive industry.
There are numerous areas where this technology could be applied such as a reliable and transparent tool for collecting, storing, managing, and using car data in a more efficient way, safe storage of the data used by intelligent navigation systems, improve security, traceability, and immutability of smart manufacturing shared data, ownership verification, and counterfeit protection.
Transfer of Personal Driving Styles to Autonomous Vehicles
A virtual environment was created to correspond to the real route in which an autonomous vehicle was modeled and the data on the previously established driving style (instantaneous speeds vs. positions) were transferred.
The influence of the Advanced Emergency Braking System in critical scenarios for autonomous vehicles
The Autonomous Emergency Braking System (AEBS) is one of the ADAS that gathered more attention in the last 10 years. The AEBS mounted on automobiles use radar (radio detection and ranging), lidar (light detection and ranging) and camera altogether, a combination of two of these or a single detection system.
The current study involves the capabilities of Simcenter Prescan and Matlab Simulink software to model and simulate autonomous vehicles’ safety system architectures which are then placed in various critical traffic situations.
The studied critical scenario consists mainly of an automated vehicle and a vehicle considered as an obstacle. The autonomous vehicle has mounted a short-range radar (SRR) and a long-range radar (LRR). Data read by the two radars is sent to an integrated solution for ADAS data collection, used further by the autonomous emergency braking system to detect potential dangers
Simulation of pedestrian collision avoidance using the Pedestrian Protection System (PPS)
The current study presents an automated vehicle that avoids a collision with a pedestrian. The main focus is on the effectiveness of the pedestrian protection system (PPS) with the main purpose of assuring the avoidance of collision with the pedestrian and contributing to a highly safe automated system traffic environment.
The simulations have been carried out by using Simcenter Prescan and one of Matlab’s modules, Simulink, in order to design and simulate autonomous vehicles which were exposed to critical safety scenarios and analyzed from a safety point of view.
In the critical scenario, the vehicle runs at various speeds (30 km/h, 60 km/h, 90 km/h) and encounters a pedestrian crossing the street. The vehicle is analyzing the object and determines if the object is a pedestrian using the Pedestrian Protection System. At 30 km/h and 60 km/h, the vehicle is able to entirely stop in safety conditions and assures the avoidance of collision with the pedestrian proving that the TTC (time to collision) value is properly selected to assuring safety conditions even in unpredictable events like a pedestrian encounter. At 90 km/h, the PPS is not able to determine the classification of the object as a pedestrian in the defined TTC in order to avoid the pedestrian, the vehicle does stop but too late to avoid the collision. This proves that for a higher speed of the vehicle, the PPS needs to enlarge the time of the TTC since the reaction time needs to be bigger when the speed is higher.
The driver-in-the-loop simulator architecture and the driving scenario designer tools
Hardware:
- MOOG Motion System 6DOF 2000E hexapod platform with UDP packet communication interface
- Display for tracking the motion platform,
- Driverțs seat mounted on the motion platform with a seatbelt,
- Logitech G29 steering wheel and pedals,
- Three HD monitors,
- Speedgoat real-time computing platform,
- Display for tracking real-time platform operation,
- ASUS Rog Zephyrus G15 PC with Nvidia Geforce RTX graphics card,
Software
Two applications: one running on the real-time platform and one for GUI and communication. These two applications were designed with the help of several successive developments in various software:
- Simulation environment developed in RoadRunner 2022a, including road geometry and infrastructure elements (road markings, buildings, traffic signs, trees, etc.)
- Rendering the scene in Epic Games Unreal Engine 4.26,
- Compiling the scene into an executable with Visual Studio 2019, using C++,
- Simulation scene with other vehicles developed in Mathworks Matlab 2022a with the Driving Scenario Designer app.
- The complete electric vehicle model, developed in Mathworks Simulink 2022a, with elements from various Matlab toolboxes,
- Calculation of the necessary displacements of the hexapod to reproduce the driving sensations, in the Simulink model,
- The transfer of hexapod coordinates through UDP packets from the real-time machine,
- Transferring pedal position and steering wheel rotation angle from the Logitech pedals to the electric vehicle model in Simulink,
- Calculation of the current position of the vehicle and its wheels to be rendered in the scene in the simulation environment.
Rezultatele proiectului
Prin proiectul DILSimEV s-a dezvoltat un instrument avansat de simulare pentru îmbunătățirea experienței de conducere a vehiculelor și pentru creșterea siguranței vehiculelor în noua eră a mașinilor electrice semiautomate și autonome. S-a creat un simulator auto constând într-un post de conducere montat pe o platformă mobilă. Utilizatorii pot interacționa cu un model de vehicul electric care rulează în timp real folosind un volan și pedalier. Cu ajutorul mai multor programe de simulare avansată, s-au creat diverse scenarii de conducere cu grafică fotorealistă. S-au identificat scenariile critice de siguranță prin procesare video și algoritmi de inteligență artificială. Acțiunile conducătorului auto în simulatorul de conducere au fost înregistrate și evaluate. S-au creat profiluri de conducere care au fost folosite pentru controlul mișcării vehiculelor autonome. Prin acest proiect s-a propus o nouă arhitectură a sistemului de control destinat vehiculelor autonome care include și strategia de conducere umană.
Rezultatul proiectului este o nouă paradigmă de interacțiune om-mașină în care acțiunile și nevoile personale ale conducătorilor auto sunt introduse în procesul de învățare al conducerii autonome. Proiectul a reușit să introducă acțiunile conducătorului auto în procesul de luare a deciziilor de conducere semi sau complet automatizată, permițând astfel un nivel optim de intervenție în condiții critice (evitarea accidentelor și creșterea siguranței în diverse condiții), precum și pregătirea strategiilor de conducere complet automatizată ținând cont de nevoile personale a utilizatorilor.
The driving scene
The simulation set-up 