Icera team forms machine-learning ADAS startup

July 23, 2016 // By Peter Clarke
A team of executives, many who previously worked at soft modem chip firm Icera Inc., have reunited to form FiveAI Inc. (Bristol, England), a machine learning software company that aims to apply artificial intelligence to autonomous vehicles.

Stan Boland, former CEO of Neul Ltd., Icera and Element 14 Inc., is co-founder of FiveAI and its CEO. Boland founded FiveAI with Steve Allpress, CTO, and John Redmond, vice president of architecture, amongst others. The company has announced an initial funding round of $2.7 million led by Amadeus Capital Partners with Spring Partners and Notion Capital.

FiveAI intends to use a "perceive-as-it-goes" approach to automotive computer vision, thereby removing the need to build and maintain detailed 3D maps of terrain and road networks. The company wants to achieve so-called Level 5 autonomy, which describes vehicles operating safely in complex urban environments without any driver involvement whatsoever.

Stan Boland, CEO of FiveAI

To this end FiveAI is developing software that will accommodate a combination of supervised semi-supervised and unsupervised learning techniques to build a real-time accurate view of the world and guide a vehicle through it. FiveAI expects to provide its software to automotive OEMs, automotive industry suppliers, rental companies and transportation operators in both public and private spheres.

FiveAI states on its website that it will use a multilayered approach that includes: sensor fusion, perception by computer vision, agent modeling to predict the behavior of moving elements within a 3D scene and motion planning for the vehicle under control. The financing is modest compared with sums being deployed elsewhere to address ADAS but FiveAI claims it will be enough to further the development of its software and begin simulator and road testing.

Not into hardware?

There is no mention of FiveAI developing hardware or processor ICs to run its software although it is well-known that while machine learning algorithms can run on general purpose processors it runs more efficiently the more the processor platform is tailored to its mathematical and memory needs.

"Computer vision can, for the first time, be superior to human abilities and we will build such a capability," said Boland, in a statement. "But safe and