Hellewell, Joseph S.
(2020)
Control strategies with human factors for active steering.
PhD thesis, University of Nottingham.
Abstract
Autonomous road vehicles are an increasingly popular research topic due to their potential to reduce the number of road deaths and road accidents while increasing energy efficiency of the vehicle and reducing congestion. To ensure a successful implementation of driverless cars, it is necessary that the vehicle's behaviour is acceptable to its occupants; that they feel as safe and comfortable as reasonably possible. What the resulting behaviour and driving style is like is different for different people and so it is necessary that an autonomous vehicle can change its performance accordingly.
This thesis presents an investigation into how existing human factor models can be incorporated into a control framework. By including models of human preference directly into a control methodology, it is possible to study how their explicit inclusion changes vehicle behaviour. The proposed hierarchical Model Predictive Control (MPC) framework is evaluated in simulation for a constant speed, Active Steering Control (ASC) implementation. This aspect of autonomous behaviour is demonstrated as it represents a typical scenario, motorway driving, and it is illustrative of the effect that human factors can have on controller design and implementation. Published driving study data is used to create different percentile drivers, allowing for the evaluation of different driving styles. The resulting behaviour is analysed and evaluated quantifiably so that the effect that human preferences have on system performance can be understood.
Vehicle models are implemented in software and validated against established results. These models are used as both the system plant in the (MPC) formulation and as simulation models of the vehicle itself. It is demonstrated that the lower fidelity models have good agreement with more advanced, and complex, vehicle models. This allows for a simpler design process, as the vehicle-controller behaviour can be simulated much more quickly.
A standard linear (MPC) formulation is then presented and its performance is demonstrated for the active steering control task. The ability to effectively constrain the vehicle’s lateral acceleration is shown, along with MPC's flexibility, by demonstrating All-Wheel Steering (AWS) performance. The necessity of providing suitable reference trajectories is demonstrated, which leads to the development of a hierarchical control framework, which creates reference trajectories for the (ASC) to track.
This developed framework is then used to implement human factor aware driving behaviour. Human factor parameters from the literature are included directly into the MPC formulation.
It is shown that incorporating human factor models of lateral acceleration preference, solely as constraints in the MPC steering module, does not provide a large differentiation in driving styles. This is because changes only occur at edge cases in driving scenarios and the human factor derived constraint is `active'. Incorporating the models into the MPC cost function, in conjunction with constraints, increases the differences in performance by allowing for a trade-off between reference tracking of system states and lateral acceleration.
To allow for this trade-off for a range of longitudinal velocities and human acceleration tolerance, a penalty model is presented that determines the relative influence of the lateral acceleration penalty in the cost function. Although the method is attractive due to its ease of implementation, it is shown that its effectiveness is limited in scenarios where other concerns dominate, such as respecting road bound limits. Nevertheless, the models presented in this thesis, along with the results and conclusions obtained from them, provide a demonstration of how human factor issues can be included explicitly into a control framework for autonomous driving.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[img]](/style/images/fileicons/application_pdf.png)