Resumen de Control and communication systems for automated vehicles cooperation and coordination
Recently, the technological advances in the Intelligent Transportation Systems (ITS) are exponentially improving over the last century. The objective is to provide intelligent and innovative services for the different modes of transportation, towards a better, safer, coordinated and smarter transport networks. The ITS focus is divided into two main categories; the first is to improve existing components of the transport networks, while the second is to develop intelligent vehicles which facilitate the transportation process. During the last couple of decades, the majority of car manufacturers and research institutes presented different forms of vehicles with Advanced Driving Assistance Systems (ADAS) capabilities. This technological advance leads to thousands of cars today in the streets, which are capable of reaching the second level of automation, with both lateral and longitudinal control of the vehicle in defined use cases, taking into consideration that the driver has to monitor the system at all times and be ready to take control. However the technological advances did not stop there, there are many vehicles on the road aiming towards the third automation level. Different research efforts have been exerted to tackle various aspects the fields of the automated vehicles research. For that reason, it is proven that the necessity of intelligent vehicle is crucial, to lower the number of accidents, traffic congestion, air pollution and improve quality of life. In addition to the more intelligent vehicles are introduced to the roads, the more cooperation and coordination among them is required. Therefore, it is of a great importance to research, develop and propose an efficient approach in order to solve the multiple automated vehicles cooperation problem.
Accordingly, this thesis is addressing the problem of multiple automated vehicles cooperation and coordination, by proposing an approach in order to solve it through various stages. The first stage is the study of simulation environments since it is essential to understand how the vehicle is going to behave in simulation before deployment in the real roads. Therefore, a powerful 3D simulator is designed for simulation of any type of ground vehicle with a various number of on-board devices. The 3DCoAutoSim simulator is developed using Unity game engine, intenerating it to Robot Operating System (ROS) framework and combine it with the Simulation of Urban Mobility (SUMO) traffic simulator. 3DCoAutoSim is an abbreviation for "3D Simulator for Cooperative ADAS and Automated Vehicles Simulator". The simulator was tested and analyzed under different circumstances and conditions, afterward, it was validated through carrying-out several controlled experiments and compare the results against their counter reality experiments. The obtained results showed the efficiency of the simulator to handle different situations, emulating real-world cars. The most important advantage for this simulator, that it can be used in any environment based on data from Open Street Maps (OSM), and using any vehicle as the driving car. This ensures the adaptability of a simulator to be used by researchers around the world to test their algorithm before deployment in real cars. Future work for the first task is to extend the networking feature of the proposed 3DCoAutoSim driving simulator to allow multiple user-controlled vehicles in the environment along with automated vehicles. Furthermore, carry out experiments for the intelligent vehicles approaches with guaranteed ground-truth.
The second stage is to use different platforms for testing the proposed approaches to automation. Therefore, small mobile robots were selected to carry out several indoor experiments of both cooperation and coordination algorithms. The selected platform was TurtleBot3 Burger, which is small differential motion mobile robot, which is equipped with optical encoders on the DC motors for the ego-motion and 360o laser distance sensor for the environment perception. Since TurtleBot3 is designed for ROS-based architectures, several available ROS packages are available in order to implement localization, mapping, perception, planning, and navigation systems. Therefore, the work in this thesis focuses on how to coordinate with multiple instances of the TurtleBot3 platforms, in order to validate the proposed cooperation and coordination architecture. Several experiments were carried-out in the GAZEBO simulated environment. The obtained results validate the viability of the proposed approach in its integration with ROS-based platforms, before the deployment of the automated vehicle platforms.
Moreover, an unmanned aerial vehicle platform was selected to test the proposed planning approach. Therefore, the SkyOnyx platform was selected, which is a carbon fiber hexacopter of total weight 4.5kg. Its controller is based on Pixhawk 2 autopilot and it is equipped with three main navigation sensors; GPS, IMU, and barometer. In addition to two on-board SJCAM SJ4000 monocular cameras. All the processing is performed on-board by an embedded computer, which utilizes ROS-based software architecture. The platform has stable flight control systems, therefore the work in this thesis focuses on the heterogeneity of the platform compared to the automated vehicle platforms for testing the planning approach. Accordingly, four different indoors scenarios were tested to evaluate the performance of the proposed approaches on the platform. The scenarios were implemented in an indoor field of 9×18 meters, using occupancy grid maps, the path planning approach estimated the obstacle-free path. The purpose of the carried-out experiments was the verification of the generic aspect of the proposed approaches and their functionality on heterogeneous platforms.
The third stage is the main focus of this thesis, which is the full development of a highly automated intelligent vehicle. The Intelligent Campus Automobile (iCab) project consists of two electric golf carts, where they were modified electrically, mechanically and electronically to achieve conditional automation and reach one step closer to the high automation level. First, in order to design the vehicle steer by wire system, the steering wheel was replaced with an actuator and absolute encoder systems. Then, in order to design the vehicle drive by wire system, the throttle pedal was deactivated and the control is applied directly to the traction brushed DC motor. Finally, in order to design the brake by wire system, the brake pedal action must be activated by an external electronic actuation, however maintaining the brake pedal functionality, as an external safety option for on-board passengers. Moreover, multiple devices and sensors were installed on the platforms, to acquire data for the environmental analysis and understanding. The devices range from two embedded computers for the ROS-based architecture, multiple sensors such as laser rangefinder, stereo camera, Lidar, GPS, compass, Kinect and ultrasonic sensors. Furthermore, auxiliary devices such as inward and outwards screens, router for the communication, warning light, warning buzzer and speakers for the on-board passengers. The iCab software architecture consists of five main layers; acquisition, processing, decision, control, and actuation. Each layer consists of several packages to send and receive data and connects the whole architecture as a single entity.
First, the acquisition layer, where the drivers packages provide a software interface to the on-board sensors, enabling the operating system to access the device functions and publish the data in the ROS messages format to the second layer, the processing layer. The processing layer holds the core systems for vehicle automation, where multiple localization systems are implemented for the estimation of the vehicle position and orientation in the environment, combined with mapping systems for the environment understanding as both globally and locally. In addition to the perception, systems are implemented in the processing layer for the obstacles detection and classification. Moreover, several communication schemes are implemented, which are utilized in the cooperation and coordination systems. Upon finalizing the processing of all data from the sensors, the decisions are made in the third layer, where the global and local planning systems are implemented. Afterward, the control layer takes the planning output and estimate the necessary actions for the vehicle movement, in both lateral and longitudinal. Finally, the output of the control layer is connected to the actuation layer, which has the steer by wire, drive by wire, and brake by wire systems for the low-level control. Hundreds of experiments were carried-out for the validation of each system in the iCab platform. Results proved the functionality of the iCab platform to self-drive from one point to another with minimal human intervention. Future work for this task is to extend the proposed approaches of the iCab platforms to the different type of automated vehicles. Accordingly, all the proposed systems will be adapted to the new automated car and later to be tested in real urban and highway environments.
For the cooperative driving goal, multiple communication schemes were introduced. The iCab platforms have three categories of communication schemes; the Vehicle-to-Vehicle (V2V) communication schemes, which share the relevant data among the vehicles in the system. Furthermore, the Vehicle-to-Pedestrians (V2P) and Pedestrians-to-Vehicles (P2V) communication schemes, which enhance the environment perception for pedestrian detection and improve the road safety, especially for the out of the line of sight pedestrians. Last but not least, the Vehicle-to-Infrastructure (V2I) and Infrastructure-to-Vehicle (I2V) communication schemes, which are essential for cities in many aspects, for instance, urban traffic coordination, traffic light cycle details, potential road hazards alerts, and other contextual information. All the proposed communication schemes were developed to operate under a Virtual Private Network (VPN), to ensure constant connectivity in the off-road environment and not only when they are in a short range of each other, in addition to the security layer that the VPN adds to the communication schemes. Future work of this task includes extending the proposed V2X communication schemes to adapt to the environment networks availability and utilize 5G networks. Therefore, the communication speed is increased among all road entities, and provide the possibility of near real-time information sharing and more efficient communication systems. In addition to, the ability to share the higher bandwidth of data among all road users.
Furthermore, through the availability of multiple instances of the iCab platform, a study of the platooning approach use case of cooperative driving was discussed and validated. The approach was analyzed using the automated vehicles and road users in an off-road urban environment through a number of experiments. The obtained results from the use of Vehicle-to-Everything (V2X) communication schemes along with a dynamic platooning approach obtained better results than vehicles that depend only on its own sensors for the environment perception. Future work for this task includes eliminating the assumption the vehicle platooning approach, which was that all vehicles have to follow the same trajectory during the whole trip. For a more generic approach, protocols of joining and leaving the platoon are to be investigated and with multiple vehicles in a distributed manner, thus each vehicle has a local leader to follow.
The next objective was to design an architecture for the multiple automated vehicle coordination as Multi-robot Task Allocation (MRTA) problem, focusing on solving the shared mobility-on-demand multiple transportation requests problem. The objective of the examined system is to collect any number of user transportation requests and allocate them to the available multiple automated vehicles. The MRTA problem governs the coordination and cooperation among the vehicles and the allocation of the requests for an optimal solution. The architecture consists of three types of entities for the process execution; the web/mobile application, the dynamically selected leader vehicle, and all vehicles in the system. For the web/mobile application, it is managed externally and independent from the proposed approach, it governs the handling of acquiring requests from the users and shows the status of the requests based on the allocation algorithm results. For the leader vehicle, it is dynamically selected based on optimized selection criteria, and it is responsible for communicating with the webserver and the executing the allocation algorithm. Finally, for all the vehicles in the system, including the leader vehicle, they subscribe to the final allocation solution and start the execution of the assigned tasks. The architecture was designed to be in a generic manner, to allow heterogeneity, scalability, adaptability, and integrability to various platforms. Furthermore, it was developed in the ROS framework to be able to test it in both simulation and real-world experiments alike. Consequently, the system architecture has been designed with two key features in mind: First, the integration with the requests acquisition systems and the vehicles must be easy to perform, as long as those systems follow certain predefined rules. Second, the implementation to solve the MRTA optimization problem must be exchangeable in the algorithm. This will allow developers to focus on the problem-solving algorithm, without investing time in the software infrastructure needed to make it work in the vehicles. Future work for this task extends the proposed MRTA cooperation architecture heterogeneity, scalability, and adaptability features to carry-out more experiments that include transportation requests for both passengers and packages, using iCab ground platforms and SkyOnyx aerial platforms respectively. Furthermore, the experiments should be carried-out in different off-road unknown environments. Last but not least, in this thesis, the complex tasks decomposition was executed analytically based on the number of passengers for each request and vehicle capacity. However, for the aim of complete autonomy of the system, further research to propose an algorithm that is responsible for the decomposition of the complex tasks based on the objective function is required. This ensures the adaptability of the proposed architecture for heterogeneous vehicles and/or tasks.
Last but not least, in order to test the proposed MRTA architecture, a hybrid optimization-based algorithm was proposed as the solution method to the MRTA problem formulated as multiple Travelling Salesmen Problem (mTSP). The hybrid approach is based on both trajectory-based optimization and population-based optimization techniques, therefore, the approach utilizes both Simulated Annealing (SA) and Genetic Algorithm (GA) approaches, in order to allocate the vehicles in a near-optimal solution, optimizing the distance covered, execution time, waiting time and vehicle energy. In order to test the proposed approach, several scenarios are selected in both simulation and real-world. The simulation scenarios are selected from well-known benchmarks of mTSP, this is in order to have the optimal cost available for a comparative study. On the other hand, the real-world scenario was designed in a way to evaluate the functionality of the proposed solution and the architecture in the platforms. The scenario involved three users using the application to create transportation requests. The three users had different starting points and different destinations, moreover the request time was close to each other to evaluate how the vehicles are going to respond. The results showed high performance of the proposed algorithm in tackling several forms of the MRTA problem and its scalability in handling multiple vehicles, outperforming all other approaches. These results show that the proposed approached was able to converge to the near-optimal values in much less computational time, which can actually be considered running in real time. For the MinMax costs, the deviation errors to the optimal costs indicated that all errors are less than 10%. This proves that the proposed approach is more capable of handling multiple vehicles and obtain more accurate solutions in all tested benchmarks.
In conclusion, the thesis main contribution is to present control and communication systems for multiple automated vehicles in order to solve the cooperation and coordination problems. The systems were tested in several simulations and real-world experiments for validation of the proposed work, ensuring its functionality and extendibility to heterogeneous vehicles. Finally, the work in this thesis derived 34 peer-reviewed publications in books, journals, and conferences, including 2 best paper awards.
