Daniel Adelberger, Luigi Del Re,
"A two-layer predictive emergency steering and escape assistant"
: 2020 American Control Conference (ACC), Serie 2020 American Control Conference (ACC), 7-2020
A two-layer predictive emergency steering and escape assistant
Sprache des Titels:
2020 American Control Conference (ACC)
Safety is a key requirement in vehicle design. Passive safety systems have been improved over decades and have been able to reduce enormously the severity of road accidents. Driver assistance systems have become more and more important in recent years, safety being one of the main goals, and in the case of an immediate collision danger some functions, like automated braking, further reduce the likelihood or severity of impacts. As a next step, automated steering will be added, i.e. the vehicle is expected not only to brake, but also to look for a safer or less dangerous alternative path. The corresponding control problem is enormously more complex, because it includes a navigation part. In this paper, we propose to state it in terms of optimal control with the task to follow a given route ? the default trajectory ? under normal conditions but to switch to a safer one once the Time-To-Collision (TTC) goes below a pre-de?ned threshold. To achieve this, we propose a two-layer structure, basically re?ecting the navigation and control tasks. The upper layer observes the traf?c participants, precomputes alternative trajectories, and determines the TTC for each of these alternatives. If the TTC along the default trajectory falls under the threshold, the safest lane is then selected as the reference. The next layer is then responsible to track the reference by NMPC, as the dynamics of the vehicle in a critical case cannot be treated as linear. A physics-based prediction method is used to determine possible future positions of surrounding road users. This prediction method is evaluated using data recorded at a junction to guarantee suf?cient performance. In the end of this work various scenarios are presented to visualize the behaviour of the proposed assistant.