TECHNOLOGY

Ex-Google engineers improve AI training for autonomous vehicles

Silicon Valley-based start-up Deepen AI’s latest technology release can help autonomous mining fleets generate accurate and scalable 4-D semantic segmentation data, meaning it can interpret data about static environments and moving objects around vehicles in four dimensions

Deepen’s new tool could speed up the annotation and validation processes of training AI data to recognise optimal areas for mining

Deepen’s new tool could speed up the annotation and validation processes of training AI data to recognise optimal areas for mining

Founded in 2017 by former Google engineers and product managers, Deepen AI provides artificial intelligence and annotation tools for autonomous systems. It has now launched its latest development in 4-D semantic segmentation for LiDAR and fused sensor data.

The company claimed that the ability to generate the accurate, scalable 4-D segmentation data (or in other words, 3-D frames progressing through time) is an industry-first.

LiDAR sensor readings have usually been semantically segmented on a non-sequential, frame-by-frame basis, to train autonomous applications. To semantically segment something means to correctly divide up objects and surfaces to identify what they are, the company explained.

"The process is tedious and costly, as every frame has hundreds of thousands of points that need to be scrubbed individually to accurately identify objects in a given frame," it added.

"For this reason, many autonomous manufacturers have had to rely on bounding boxes, which are easier to produce but are far less effective at discerning edges between objects, like a road and a sidewalk. "

Deepen believes its LiDAR technology can provide more precise readings for identifying and semantically segmenting objects in four dimensions.

The technology can also be used in mining, where dangerous tasks are increasingly being operated remotely or fully automated.

Mohammad Musa, founder and CEO of Deepen, said: "This is a massive step for developing AVs [autonomous vehicles] safe enough to deploy on an enterprise scale.

"Furthermore, a massive door has opened that will give innovators the opportunity to unleash their creativity to leverage 4-D data segmentation - perhaps in ways that we haven't even been thought of before."

The new technology is now available to Deepen's customers as a limited-access beta.

A growing series of reports, each focused on a key discussion point for the mining sector, brought to you by the Mining Magazine Intelligence team.

A growing series of reports, each focused on a key discussion point for the mining sector, brought to you by the Mining Magazine Intelligence team.

editions

Mining Magazine Intelligence Future Fleets Report 2024

The report paints a picture of the equipment landscape and includes detailed profiles of mines that are employing these fleets

editions

Mining Magazine Intelligence Digitalisation Report 2023

An in-depth review of operations that use digitalisation technology to drive improvements across all areas of mining production

editions

Mining Magazine Intelligence Automation Report 2023

An in-depth review of operations using autonomous solutions in every region and sector, including analysis of the factors driving investment decisions

editions

Mining Magazine Intelligence Exploration Report 2023 (feat. Opaxe data)

A comprehensive review of current exploration rates, trending exploration technologies, a ranking of top drill intercepts and a catalog of 2022 Initial Resource Estimates and recent discovery successes.