Route Risk Assessment - A Smarter Way for Safely Deploying Autonomous Vehicles
- nkc
- May 20
- 3 min read
The SCALE project will see a fleet of three self-driving shuttles ferry passengers along a 7km route from Birmingham International Rail Station to the National Exhibition Centre and Birmingham Business Park.
The Connector project will deploy three 52-seater buses at the Cambridge Addenbrooks Hospital and at Cambridge University’s west campus.
Conode has identified the need to not only understand the risk landscape around these sort of planned route-based AV deployments, but to be able to also develop a robust safety plan. That's why we've added new computer vision capabilities to our tool, making it easier for safety managers to analyze risks for such AV deployments. Integrating the new GenAI model has accelerated image and video data processing for identifying road infrastructure hazards.
The Featurizer
Driving this methodology is our powerful agent, fondly called the Featurizer. Capable of interpreting and extracting relevant data from visual inputs, it has helped us and our consortium partners obtain road features easily. Here's how it works:
🖼️ Simply upload all your images onto Conode. These could come from static CCTV cameras, drones, dash cams, or even photos taken manually during site visits.
💬 Prompt the Featurizer on what you’d like to assess. Whether its pedestrian crossings, bollards, a particular type of road markings, or any other infrastructure-related features, you name it!
💡Evaluate road risk. Based on an assessment of other analysed routes and similar road features, you can determine which features contribute to a higher risk of a collision.
As a first pass, we can ask the Featurizer to identify general hazards along the route. It will look through all image data and annotate them with for example, 'sharp turns', 'cracked road surfaces' and 'pedestrian crossing areas' where present. Then, we can additionally request to identify specific environmental features that could be related to road risks, such as, 'hedge rows' or 'bollards'.
Once the planned deployment route has been annotated, we can then compare the various risk features against other locations where AVs are already operating on. This extra information allows us to generate a risk profile on how AVs can handle potentially dangerous situations.
Fusing Together Datasets for a Comprehensive Risk Assessment
At the heart of the Conode approach is a bunch of AI powered agents capable of transforming the AV industry. Our method takes advantage of previous assessments of roads where AVs have driven — even internationally, like in the US where AV trials have scaled up rapidly. The ability to analyse road risk purely from images is a game-changer because it is not reliant on the availability of HD maps.
Featurize your Image Data
If you're interested in leveraging GenAI-powered analysis tools to similarly build a safety case for your data, hit that button and contact us to learn more!
Comments