AbstractAutonomous vehicles are fast becoming one of the major areas of research for automotive engineering. When the vehicle is fully in control of its actions, this raises questions concerning automotive safety, such as what an autonomous vehicle will do when faced with an imminent collision. Assuming that autonomous vehicles cannot crash is a dangerous over-estimation, and so systems need to be in place to limit the potential risks to vehicle occupants.
The aim of this research is to develop a control strategy for an autonomous vehicle, called the Host Vehicle, to avoid or mitigate collisions on longitudinal multiple carriageway roads. These multiple carriageway roads are arterial roads, called motorways. The potential collisions must be assessed from the perspectives of all vehicles involved in the impact, to prevent a selfish decision being made by the Host Vehicle. The main scenario of this thesis is of an autonomous vehicle driving on a three-lane motorway, with potential collisions in each lane. Therefore, each lane is a possible choice for the autonomous vehicle to select. This thesis proposes a system that selects the safest possible crash for an autonomous vehicle when faced with multiple possible collisions. This system aims to avoid or mitigate potential collisions.
This system requires expertise from several different areas. Autonomous highway platooning systems have been developed and tested to demonstrate that autonomous vehicles can crash. If a potential lane-change manoeuvre is required to avoid or mitigate a collision, the manoeuvre must be planned and assessed to ensure the risk to the autonomous vehicle safety is not increased. The potential collisions must be assessed for severity, requiring modelling of these collisions to produce metrics for a decision-making process. The severity of a collision is greatly influenced by the impact velocity, which therefore requires the impact velocity of the potential collisions to be simulated. Two simulators are developed for the case when Vehicle-to-Vehicle communication is available, and the case when this communication is not available. These two cases influence the available parameters for calculating the severity of the collision. Once all the required outputs from the simulators and modelling are produced, describing the potential collision severity of each available lane, these are used to select the lane with the least severe collision scenario. Multi-Attribute Decision Making (MADM) is used to assess the outputs from the simulators, and make the decision of which lane the autonomous vehicle should drive into. MADM has not been applied to this type of problem before. The MADM methods which are investigated for this research problem are the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), the Analytical Hierarchy Process (AHP), and the Analytical Network Process (ANP).
The novelty of the proposed system includes the application of Multi-Attribute Decision Making to select the least severe collisions for the Host Vehicle to take, therefore an autonomous decision is made to improve the safety and survivability of vehicle occupants. This system is supported by the development of a new two stage collision modelling to describe the severity of multiple vehicle collisions. These two stages are two separate rear-end collisions, the Host Vehicle collides into a vehicle ahead, and a vehicle behind collides into the Host Vehicle. A steering and braking trajectory planner is developed to give the Host Vehicle multiple actions to select from.
The proposed simulation methods are tested and evaluated with respect to the decisions they make. This includes simulating several scenarios in which the simulated vehicles vary their behaviours. The result is a recommended lane choice, so the car decides on the safest collisions to have. The scenarios vary the input parameters such as initial velocity, available headway distance and braking of vehicles. Each scenario is tested, and the lane selection is presented. The parameters are varied in a sensitivity analysis to demonstrate how the lane selection can change based on the inputted scenario. The TOPSIS and AHP methods generated good decisions in line with the decisions made by the subject expert. The ANP method would require further parameter tuning. The proposed system is intended to be an evolution of the current Adaptive Cruise Control and Collision Avoidance/Warning systems including Automatic Emergency Braking.
|Date of Award||26 Mar 2019|
|Sponsors||Engineering and Physical Sciences Research Council & Jaguar Land Rover|
|Supervisor||Dobrila Petrovic (Supervisor), Kevin Warwick (Supervisor) & James Pickering (Supervisor)|