Dev board for insect-like visual AI

February 08, 2021 // By Peter Clarke
Dev board for insect-like visual AI
Opteran Technologies Ltd. (Sheffield, England) is offering a development kit (ODK-2) for its silicon-based approach to deep learning using optical flow techniques.

The company claims it has a different approach to artificial intelligence based on studies of insect brains. But applying insect-style learning the company expects to enable machines to see, sense, decide and navigate.

The kit integrates the Opteran-See technology to provide stabilized images and the Opteran-Sense technology, which calculates the "optical flow" across the field of vision. This optical flow includes intrinsic potential collision signals – such as an object getting larger in the image field – that help a device take course-corrective action.

The company claims this "insect" approach reduces the size and cost of an AI approach to autonomous navigation while increasing its robustness. Insects require only small amounts of data to navigate, avoid obstacles and interpret their environment.

The ODK-2 measures 4cm by 5cm by 1cm, weighs 30g and draws a few watts of power. There is no need for extensive power-hungry pre-training or pattern matching. It can be plugged in with a USB power cable or similar interfaces, so it can be quickly up and running deployed at the edge; and with an incredibly robust signal it ensures reliability.

Opteran envisages the use of ODK-2 as collision-avoidance back up to a human operator of a vehicle.

ODK-2 is available and will shortly receive a firmware upgrade to include Opteran Direct in support of simultaneous localization and mapping (SLAM).

Related links and articles:

News articles:

Optical flow AI startup Opteran raises £2.1m

Startup raises €5m for neuromorphic edge chip

Aspinity raises funds for analog ML roadmap, expansion

GrAI Matter raises $14 million for SoC roll out

Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.