A Holistic Decision-Making Framework for Integrated Safety
Mohsen Nosratinia
Department of Signals and Systems,
Chalmers University of Technology,
Go¨teborg, Sweden
nosratin@chalmers.se
Henrik Lind, Stina Carlsson
Department of Safety Electronics
Volvo Car Corporation
Go¨teborg, Sweden
{hlind1,scarls12}@volvocars.com
Niklas Mellega˚rd
Department of Applied IT,
Chalmers Tekniska Ho¨gskola,
Go¨teborg, Sweden
niklas.mellegard@ituniv.se
Abstract
A Bayesian decision-theoretic decision-making
framework for integrated vehicle safety systems is in-
troduced. The framework tries to address the increas-
ing need for introduction of optimal decision-making
to integrated vehicle safety. The framework tries to
capture all the interdependencies between systems in
one optimisation problem by designing appropriate risk
functions. This is achieved by incorporating driver be-
haviour model and pre-crash occupant position track-
ing. New software methods and tools should also be
developed to efficiently accommodate this. The frame-
work, in general, leads to higher design flexibility and
scalability 1.
1. Introduction
The last 15 years have seen a dramatic increase
in the amount of software controlled functions in ve-
hicles. By using computers, sophisticated active or pri-
mary safety functions can be created, e.g. functions
that warns a driver if approaching an obstacle or au-
tonomously applies brakes if the vehicle is about to hit
a pedestrian. Such functions have started to become
commonplace in the premium vehicle segment. How-
ever, today, these primary safety functions are still im-
plemented largely as stand-alone systems with little or
no cooperation in order to make decisions about the op-
timal response to a potential incident. This severely lim-
its the sophistication of the actions the vehicle is able to
take. Although, there are ongoing research initiatives
that aims at integrating functionality common to many
function, e.g. SEFS which integrates the vehicle sensors
in order to create an improved perception of the vehicle
1This research is partly funded by The Swedish Governmental
Agency for Innovation Systems (VINNOVA) as part of Intelligent Ve-
hicle Safety Systems (IVSS) programme.
surroundings that multiple vehicle functions are able to
access, INSAFES subproject within PReVENT [1]. In
order to further the vision of an integrated vehicle safety
system, there are a number of challenges that need to be
addressed, such as
• Finding an optimal scheme for decision-making
• Understanding how the position of vehicle occu-
pants changes during the pre-crash phase of an ac-
cident
• Understanding how a driver reacts to autonomous
vehicle actions and how a system may integrate
these in its decisions
• Methods and tools for efficient system develop-
ment
In this paper we outline a research project called
ASIS (Algorithms and Software for Improved Safety)
aimed at investigating the future of vehicle safety sys-
tems. Our primary research goal is to propose a new
method for in-vehicle decision making in critical traf-
fic situations, based on all available information in the
car. By employing an optimization-based approach that
tries to capture the effects of decisions on occupants and
drivers behaviour while meeting requirements on false
activations and occupant injury risk in a crash, a set se-
quence of actions is decided. These actions could in-
clude using up to all actuators available in the car. We
do this by addressing the following research questions:
1. What properties should cost function have to en-
sure specific criteria? How should cost functions
be designed to handle trade-off between different
requirements?
2. How to model the driver incorporating individual
and scenario based variations to support effective
interventions?
3. How to model and predict occupant injury risk de-
pending on the scenario and occupant position?
2010 IEEE Intelligent Vehicles Symposium
University of California, San Diego, CA, USA
June 21-24, 2010
ThB1.1
978-1-4244-7868-2/10/$26.00 ©2010 IEEE 1028