From Perception to a Framework for Verification of Autonomous Behaviour

About this session

Perception supplies autonomous systems such as self-driving cars or unmanned aerial vehicles with inputs about the status of the system as well as the surroundings. These inputs are then combined with other information and used to plan, control and coordinate the behaviour of an autonomous system. Inputs from perception are also needed to verify whether the system behaves correctly. Correct behaviour is determined by the manufacturer, but also legal regulations such as road or air traffic acts. Such requirements are complex, and can greatly differ. To capture correct behaviour, we translate requirements and legal regulations into machine readable logic language constraints, which can be verified at runtime.

In this talk, we show the path from perception to the verification framework and illustrate the ideas on some examples. By highlighting some of the challenges, we describe the requirements on perception and outline an interface between perception and verification.