Skip to main content

Decisions and Behaviors for Cognitive Automobiles Research

Final Report Summary - DBCAR (Decisions and Behaviors for Cognitive Automobiles Research)


The research presented herein belongs to the Intelligent Transport Systems and Services (ITS) field. The proposed work is to design behaviors and decisions for Autonomous Vehicles (vehicles that drive by itself, also known as the "driverless cars"). After a correct interpretation of a traffic situation, the most appropriate behavior will be triggered, a path to follow will be designed, and the proper orders will be sent to the car actuators (steering wheel, throttle and brake). This work is focused in the cooperation and coordination among several autonomous vehicles. Specifically, situations where the planned trajectories of the vehicles interfere among them. The project objective is to propose a system that will allow the vehicles to coordinate their trajectories. This trajectory coordination keeps a safe distance between the vehicles at every moment, and increase the flow of vehicles in certain situations, like intersections and roundabouts.

1. Work carried out to achieve the project's objectives:
After learning how the software development platform work, and how the control software and the simulation software works, an extensive state of the art search was conducted. A solution where the autonomous vehicles works as agents and are coordinated from a central agent was selected. Two software modules were developed to interact with the simulator and the control software without interfering with the rest of the team work. A module collects the needed information from the vehicle database and send it to the external control software. And the second module was also designed to overcome the normal outputs of the vehicle controller. With the information of all the vehicles within the scene, a cooperative trajectory for each vehicle was designed and sent to them. The first tests and conclusions where carried out for two vehicles cooperation. Based on this experience, the program was modified to allow any number of vehicles.

2. Results
The two main tasks proposed in the project were: to optimize the expected flow of vehicles at a high safety level by preplanning the trajectories of cooperative autonomous vehicles at intersections; and to continuously monitor and adjust the trajectories of the cooperative autonomous vehicles with respect to safety. In order to fulfill the project tasks, a centralized join trajectory design unit has been implemented. This unit collect the trajectories of all the cooperative autonomous vehicles and send them the new trajectories that ensure an optimal flow of vehicles keeping the safety level at maximum.

Several techniques has been tested within the centralized join trajectory design unit. As the unit collects and modifies 5 seconds of each vehicle trajectory sampled each 0.33 seconds, the different control actions taken in each point for each vehicle makes the exhaustive optimal search techniques useless, even with only 2 vehicles. So other non-complete optimal search techniques has been used. The particle swarm optimization (pso) method has been chosen because 2 decisive factors. First, the human a priori knowledge of how to solve this situations can be used to set a good initial particle population. And, secondly, the pso convergence speed guarantees good results in the given frame time.

In order to create the centralized join trajectory design unit, two modules has been added to the control software of each vehicle. One of them collects the state of the vehicle and its trajectory, and send them to the unit. And the second receive the new trajectory from the unit and overrides the current trajectory of the vehicle with the new one. These two modules, in combination with the vehicle control program that is able to be used in simulation, makes possible to run multiple iterations of the simulation and test different coordination algorithms. These modules create an excellent testing ground for further coordination tests.

The collaboration between the host center and other european research centers has been increased. Four different spanish research centers have sent PhD students to the KIT in this two year period, and is expected that this cooperation will continue in the future. Also, european projects has been asked for in cooperation with other european research centers. In the CAVE-UP project proposal in 2013, a method to automatically collect high precision digital maps to make easier the autonomous driving was proposed. And taking adventage of the needed infraestructure, a method of autonomous vehicle coordination using the idea of a centralized join trajectory design unit was also included. The proposal do not succeed, but a new proposal has been prepared for 2015.

The researcher has got the opportunity to learn how the experimentation vehicles of the KIT work, to extend his knowledge in several autonomous vehicles related areas, to learn the KIT work methodology, and to learn the proper tools to manage large research groups. The experience has also allow the researcher to be better known in this field and to meet people from the automotive industry and other research centers.

3. Conclusions
An exhaustive method to explore all the possible combinations is unfeasible. Alternative methods for the optimal search with a high convergence speed should be used. The PSO method has proved to be very conviniant for this situation. The time increase of the PSO method once the search dimensions are increased is lineal, so the extension of this method to control more vehicles is easier than other methods like the graph based search methods.

4. Potential impact and use
The project can reduce the waiting time, and increase the flow of vehicles at intersections, roundabouts or similar situations. And also increases the safety by ensuring a minimum distance between the different vehicles who are controlled by the system. It can be applied directly to highly robotized areas like some industrial areas. But, to bring the research to practice in specific areas of a city, a larger project would be required.

5. Socio-economic impact of the project
The implementation of the project would represent significant savings in fuel and travel time, as the vehicles would reduce the time wasted waiting to cross intersections or entering in the roundabouts. Also, as the coordination between the vehicles is done based on its trajectory, wich includes the speed profile for the next seconds, the following vehicle do not need to overreact to the preceeding vehicle, and overreacting means a waste of energy.