Title: Phd position in safety argument for autonomous vehicles
Laboratories: LAAS-CNRS & IRIT, Toulouse, France
Autonomous vehicles, inspired by mobile robots, operate in uncertain environments and face many situations that cannot be fully specified or tested. In addition, there are uncertainties induced by the technologies used (e.g., for the perception of the environment) or by the presence of residual faults within these systems. These threats raise the question of the confidence placed in them, including safety. This lack of confidence can jeopardize the social acceptance of these systems and therefore their existence. Many dependability techniques are useful to meet the requirements of regulations or standards, including on failure probability calculations. These pre-established safety techniques are well adapted to known contexts, but are not applicable to the case of autonomous vehicles, partly due to a lack of exhaustive information. It is therefore necessary to provide solutions to quantify the confidence despite the presence of these uncertainties due to the incomplete nature of the information.
A technique in full expansion in the design of critical systems, is to develop safety arguments (or safety cases), which consolidate in a structured way all the evidences of safety. However, these elements, and in particular for autonomous systems, have a degree of uncertainty (related to the above-mentioned causes, or the use of necessarily inaccurate expert assessment), which must be quantified and propagated to global confidence. We propose to use the notion of imprecise probability to make this assessment. The aim of this thesis is to propose a method allowing to build, despite the uncertainties, a safety argument potentially applicable for the certification of autonomous vehicles. This thesis will be co-supervised by experts in autonomous system safety (LAAS-CNRS) and imprecise probabilities (IRIT). The method developed will be applied on a real system that will come from on-going projects on autonomous robotics.
Student profile: applicants who had training in computer science / mathematics but also open to profiles on the design and analysis of embedded systems.
Keywords : Safety argument, imprecise probability, dependability, autonomous vehicle
Starting date : January 2019
Jérémie Guiochet, email@example.com
Didier Dubois, firstname.lastname@example.org
To apply : https://app.laas.fr/boreal/web/en/voir/these/simple/without/197