British self-driving car tech start-up FiveAI picks up $2.7m in venture funding

FiveAI's AI-led approach to autonomous vehicles won't require complex 3D mapping, claims the company

A British start-up developing autonomous vehicle technology has scooped funding of $2.7m (£2m) from venture capital investors, including Amadeus Capital Partners, Spring Partners and Notion Capital.

FiveAI claims that its approach to autonomous technology will make more use of artificial intelligence and machine learning, and less complex 3D mapping. The aim is to develop a system that can negotiate all environments based on its comprehension of those environments, rather than pre-built maps containing all the likely hazards a self-driven vehicle will need to understand.

Early approaches to autonomous vehicles have required accurate, 3D prior maps built using point-cloud technology so that at run time each vehicle can correlate against the prior map to accurately localise itself and establish a track to follow.

"Developments in computer vision and AI are achieving breakthrough after breakthrough. In particular, the latest types of deep neural networks as applied to computer vision are now delivering step improvements in being able to accurately identify objects, their type, their distance from the vehicle and their exact motion," CEO and co-founder Stan Boland told Computing.

"High accuracy means many fewer false negatives (when a vehicle should have identified something and failed to do so) and fewer false positives (when a vehicle wrongly identifies a hazard and acts incorrectly). Low false negatives usually means more false positives, and vice versa; so only a better perception capability can resolve the conflict and that is an important element of what we are building.

"That better perception capability will also enable us to construct an accurate view of the scene, the fixed background (such as roads, kerbs, buildings, signs and so on) and moving objects (pedestrians, cyclists, cars), meaning that we do not need highly accurate prior maps and, thus, don't need to rely on the ability to place us precisely on that prior map to determine the exact position of obstacles, turns and signs.

"Instead, our system will perceive as it goes. That means we won't need to use LIDAR technology to construct and maintain prior 3D dense maps for every possible road we might drive down (there are 37 million kilometres of road on our planet); we can build a detailed view of the world as we go on top of a much simpler map, more like a 2D navigation map we use today. All this means it will be much safer."

This is, it ought to - if and when the technology is perfected. While the $2.7m investment will enable the company to recruit and develop its approach to autonomous vehicle technology, it will no doubt require further rounds of funding to develop it and take it to market.

However, Boland has plenty of experience in tech companies and start-ups. He was formerly the CEO of Acorn Computers towards the end of its life, and was the founder and CEO of Element 14, the digital signal processing technology spin-off from Acorn that was sold to Broadcom for $640m in 2000.

Boland also co-founded energy-efficient chip set designer Icera in 2002, which was sold to graphics technology company Nvidia in 2011 for $367m, and CEO of Internet of Things start-up Neil, which was sold to Huawei for $25m in September 2014. He founded Blueprint Technologies, which morphed into FiveAI, in last September.