LAMBDA-V: Real-life driving data to improve autonomous vehicles

A partnership between CloudMade, Aimsun, Birmingham City Council and Trakm8 is aiming to use human driver data to improve the performance and acceptability of connected and autonomous vehicles (CAVs). ‘Learning through AMBient Driving styles for Autonomous-Vehicles’ (LAMBDA-V), is a one-year feasibility study into how human driver behavior can be analysed using anonymised telematics data to accelerate the adoption of CAVs. For example, the likelihood of a human driver swerving to avoid a pothole, or how and when they apply the brakes when entering a 30mph zone. This data can help better inform the decision-making of CAVs.

LAMBDA-V is part of the UK Government’s £22m (€25m) funding from the Centre for Connected and Autonomous Vehicles (CCAV) for projects to develop autonomous vehicles. The lead partner is CloudMade, bringing expertise in machine learning and human driver behavior modeling. The other partners include Trakm8, which will collate and analyse anonymized sample data from thousands of vehicles. The other consortium partners are Birmingham City Council as the highway authority with legal powers and duties; and smart mobility software expert Aimsun.

James Brown, CTO of CloudMade, said: “Understanding human behavior and modeling this behavior is one of the key elements in humanizing autonomous vehicles and enabling personalisation of the vehicle. CloudMade, with its extensive experience and expertise in machine learning and human behavior profiling, is uniquely positioned to utilize these capabilities in this program. The CCAV grant will enable us to accelerate the development of solutions that learn individual driver behavior and derive the necessary rule-sets and approaches to modeling and adaptation during the drive.”

LAMBDA-V is a one-year study on the feasibility of processing existing massive datasets, to understand the parameters needed for modeling human drivers and how to extend them to make vehicle rules, improving current technology and modeling impact to balance comfort, capacity and safety. This could ensure CAV behaviour meets the needs of both regulators and customers.

The project will focus on innovative exploration of a full end-to-end data chain and business model in a mixed fleet environment. This integrates vehicle maker and road operator perspectives on CAV behavior; and examines how to develop privacy-law-compliant datasets for other CAV projects. It brings together those who develop CAVs and modeling software with data from massive mixed fleets of anonymised drivers across the UK, rather than small fleets of specialized vehicles in one location.

New rules for safer and more efficient driving behavior may be built from data from existing vehicles, based not just on road laws but on how humans drive vehicles in specific circumstances. These could be ‘tuned’ by modeling how CAVs and other vehicles then behave in a mixed fleet, which will help to tailor early CAV behavior to match that of human drivers and thereby improve confidence for early adopters.

The key output will be identifying potential product improvements for all partners to make data, modeling and rules generate new sales. The benefits would include: reduced unforeseen impacts on traffic; patents on rules for CAVs; an improved understanding of early mixed fleet operation of human and automated vehicles and how to make early level self-driving vehicles attractive to users; and improved understanding for highways authorities and vehicle makers regarding how to deploy CAVs on a variety of real-world roads.

Author