BBC Click features dRISK Edge Case library on Future of Transport episode
15th Jun 2021 dRISK
BBC Click, the coveted weekly BBC television program covering technology news and recent developments in the world of technology and the internet, featured dRISK.ai, our edge case knowledge graph, and the D-RISK project in its recent episode dedicated to the Future of Transport.
The work of dRISK.ai and the D-RISK consortium collecting real life traffic incidents from CCTV, front facing cameras, insurance and accident reports as well as reports submitted by the public, was highlighted as the crucial asset in ensuring Autonomous Vehicles (AVs) are safe on our roads. We collect accident data, but more importantly near miss data which is often overlooked and remains unlogged.
Importantly, these data can be used to train an AV to ensure that they are able to safely cope with these types of incidents, which we call Edge cases. We have retrained industry-standard object detection with our simulated edge cases and achieved results that confidently show up to six times faster detection of high risk events, with twice the confidence.
dRISK.ai, along with the D-RISK consortium including Imperial College London, Claytex Services and DG Cities, are currently on track to deliver the World’s first true driving test for AVs. by early 2022 as part of a project funded by the Centre for Connected and Autonomous Vehicles (CCAV) and managed by Innovate UK. dRISK.ai lead a consortium of partners made up by Claytex, Imperial College London and DG Cities all working to create a driving test for Autonomous vehicles. We hope that this can be used as a benchmark, ensuring that AVs are safe and can handle all manner of Edge cases.
Safe and reliable AVs will reduce the annual number of road fatalities and serious injuries and training on Edge case data will facilitate this.
The episode can be seen via the BBC iPlayer here: https://www.bbc.co.uk/iplayer/episode/m000x46t/click-the-future-of-transport
or on YouTube here: https://www.youtube.com/watch?v=W5liH8_Nuz0&t=234s