Grayscale AI is building neuromorphic software that integrates with existing tech to provide human-like precision and understanding to navigating the roads. It’s faster, more energy efficient and has the ability to perform well in difficult conditions.

Grayscale AI is a tech start-up dedicated to creating optimization-driven AI, with the intention of mimicking a human’s neural network. Their solution leverages neuromorphic cameras, which detect changes in the environment, a contrast to traditional cameras, which take continuous snapshots of a scene leading to an excess of redundant information and high data output. The tech stack offers a superior processing time that’s both cost and energy efficient.

It’s spearheaded by Dragos Stanciu, a software developer with a passion for neuroscience. Dragos channelled his interest into computational neuroscience, creating mathematical models to represent how the brain works, eventually applying this knowledge to robotics and neuromorphic cameras.

Dragos explains:

When we talk about neuromorphic computing, it’s a completely new computing paradigm, but there’s a huge opportunity to move technology, and the market, towards this new approach to processing data and interacting with the environment.”

Now he continues to grow the company, with the aim to add the best and brightest in computer vision, computational neuroscience and robotics to further test the limits of this cutting-edge technology and achieve increasingly more dynamic, efficient neuromorphic vision solutions.

What makes Grayscale AI special?

Grayscale AI realised that the automotive industry can benefit from a greener and smarter solution to spatially understanding environments. By using neuromorphic computing, they can create a cutting-edge solution that can dramatically outperform traditional sensors, as it has the capacity to process complex scenarios with little input needed. It has the potential to be highly efficient at detecting and avoiding hazards and promises to perform well even in the most challenging, vision-obscuring environments.

Neuromorphic chips are fundamentally different from the chips currently used in smartphones and laptops. The algorithm uses computer science to reframe classical logic-based AI into a more robust and flexible solution.

Traditionally computers split up the CPU and the RAM, with the data moving between them, however neuromorphic chips house the computation and memory in one piece of hardware.

These chips use “spiking neurons” as an element of computation. Instead of computing in a rigid binary “vocabulary” of 1s and 0s, the “neurons” compute in a more flexible manner. Also, unlike CPUs, these chips don’t have a central clock to synchronise their computing steps, which allows the “neurons” to operate without a predefined order.

In a neuromorphic camera, each pixel is independent in picking up the changes in the environment. It combines low-power AI computing, local interpretation and analysis, and achieves negligible standby consumption, making it up to 10,000 times more efficient than its counterpart. This creates a more sustainable solution that has a lower carbon footprint. It also increases the speed of performance and crucially, it’s reactive to the environment. Instead of processing information sequentially, it can emulate and learn algorithms at silicon level – all it needs is the stimuli to learn a pattern from. The sensors on the camera also have a high dynamic range which enables it to work well in low light conditions.

When applying this to the automotive industry it can be programmed to detect vulnerable road users like pedestrians and cyclists in microseconds.