Simulation-based autonomous systems testing – from automotive to defence
(Room 320B)
04 Dec 19
9:30 AM
-
10:00 AM
Tracks:
Full Schedule, Wednesday Schedule
The introduction of autonomous systems into safety-critical domains poses many challenges – from system safety assurance to public acceptance of the use of such systems. Recent autonomous vehicle incidents have underlined the need to establish clear regulations for the introduction of such systems into public environments, while the challenge of conducting adequate testing in the real world makes simulation-based testing an attractive proposition – reducing time, cost and risk associated with development and evaluation.
This paper will seek to elaborate on an XPI-led feasibility study examining the use of simulation for the certification of autonomous vehicles (CAVinSE), co-funded by the UK Centre for Connected and Autonomous Vehicles, and draw parallels with accrediting autonomous systems for defence applications. This will include consideration of the emerging standards, regulations and methodologies in the automotive sector with regard to autonomous systems. Moreover, it will highlight the importance of establishing a trusted environment for autonomous systems testing activity – with a root of trust that can be relied upon by certifying bodies, system developers and the general public. Potential approaches for accrediting simulation tools associated with such autonomous systems testing will also be addressed.
The paper will consider how the different operating environment and capability of defence systems, as well as different regulatory frameworks, could nonetheless benefit from the outputs of CAVinSE and similar endeavours in the automotive domain. As well as XPI’s work on CAVinSE, a use case for simulation-based autonomous systems testing in the maritime domain will be covered.
In concluding, some of the key technical challenges that remain in this domain will be identified – with a particular focus upon sensor models, representation of the physical environment and adequately representing human behaviours in simulation.