|GABRIELLI ALESSANDRO||Cycle: XXXII |
Section: Computer Science and Engineering
Tutor: BASCETTA LUCA
Advisor: MATTEUCCI MATTEO Major Research topic
:Improving passenger's comfort in autonomous drivingAbstract:
The research is focused on the creation of an autonomous driving system which allows a car to drive with a style that is comfortable for the passenger, i.e., the former driver. Indeed, if in the future, passengers will not feel relaxed and comfortable in autonomous driving conditions, they probably will not use them. In literature, it is possible to understand that you cannot simply acquire data on the driving style in a manual driving and then reproduce it. This is because the passengers tend to prefer a more defensive style rather than their own and, this, depend on the lack of control perceived when they are only passengers. Thus, this research, has the aim of adapting the driving style in real time measuring or estimating the stress and discomfort of the passenger, using it as a feedback of goodness of the driving style. It is possible to divide the system in two main part: the navigation one and the passenger comfort measuring. . The autonomous navigation system is able to change the driving style, modifying the constraints that the control commands must comply, and sending these generated commands to the car. The changing in the driving style is done using the passenger’s stress level coming from the passenger comfort measuring part. Those changes have the aim of decrease the passenger’s stress level that are perceived by using the physiological data.
Measuring the passenger/driver discomfort in real time is not an easy task. In this research, in order to acquire the stress of the passenger we decided to use the physiological sensors since such data can be collected continuously and are reliably. Following this, we have set up a commercial car, being able to capture physiological signals.
The navigation part uses ROS (Robot Operating System) navigation stack.
The global planner uses Search-based Planning Laboratory (SBPL) and precomputed motion primitives to create a global path, from the starting point to the goal, agreeing with the ackermann kinematic constraints.
The local planner will be able to control the car using the Model Predictive Control framework, transforming a classic controller problem into a mathematical optimization, so that it is possible to simply insert constraints and limitations to the system. One advantage of using this implementation is that it considers the dynamic of the vehicle and not only the kinematic. However, a major advantage is that with that approach we are able to add, remove or modify the constraints on the control parameters dynamically and, thus, impact autonomous driving style.