Abstract
Chess and his team at dRisk invited us along to their latest set of testing to see if self driving cars are safer than humans... by having a person step out in front of an autonomous vehicle to see whether the auto braking system will kick in or if the Tesla will hit the person. This is testing a Self-Driving car's auto braking system with a real person... self driving car vs human! Currently AV’s in full self-drive mode aren’t allowed on the UK roads and therefore any testing would need to be done on designated test tracks. By tricking the car into thinking it is driving on a real road they’re able to test simulations. This allows them to manipulate a scenario changing things such as weather, what time of day it is, the colours and textures of background materials… they can even change the clothes that the people in the simulations are wearing, and all with the intent of seeing what the AV’s AI will recognise. For example, will it see a person wearing a Zebra costume in front of a black and white background? That face in the thumbnail is me... And that look on my face is genuine shock! Little did I know when organising this video that I was going to testing the Autonomous Vehicle's (AV) auto braking system will a real person… the very person that I was about to interview. The results were quite shocking. Let’s just say it made for an interesting conversation. But... before we jump in front of a moving car, let’s understand why we’re jumping in front of a moving car. Self-driving cars have been a vision of the future for over a century but in the last 20 years, autonomous driving has moved from the realm of science fiction to reality, with many experts claiming that self-driving cars can be trained to be safer than human drivers. But when AV's arrive in their masses, how will the UK government have confidence that these are, indeed, safe? Currently, the only place testing AVs is in the USA. In the US testing is being carried out on public roads by collecting their data through driving many many miles with what Autonomous vehicle safety company dRisk believe is essentially prototype software. They believe that this is an accident waiting to happen as these vehicles have not been tested in challenging scenarios, only the basic ones clocked on the typical day to day journeys. AV manufacturers are using the public to test the software. Think of this as Microsoft releasing a new operating system and it being down to the end users to report the issues they find. dRisk have created a multi-level framework with the capability to identify realistic edge-case test-scenarios from real and simulated data, human-driven fault detection methodologies and public responses to connected autonomous vehicles (CAVs). This huge variety of data sources will be structured and fed into a comprehensive knowledge graph of all CAV risk scenarios. From there, representative test cases are drawn and fed into one of several physical and cyber- simulators which will directly test the vehicle control system. In summary dRisk are delivering the first, true driving test for the Autonomous Vehicle. The idea behind this is that currently there are two areas which give the UK government confidence in car and driver: For the Car - Type approvals for cars and car parts such as. NCAP testing, Crash testing, MOT testing etc, and for the driver - Driving test, Practical, theory, hazard perception. The premise behind dRISK is that they collect the hazardous scenarios that happen on the roads around the world from sources such as CCTV cameras, dashcam footage, Accident reports, Insurance claims… even Outreach from the public via DG Cities, and this is an important point for near misses which otherwise are lost. These are re-created in the simulation environment and are then clustered into similar types in our knowledge graph.
