Here at Tech.eu, we get a lot of press pitches from startups working in AI — primarily generative AI. I'm always looking for something different, so my interest was piqued when I came across Opteran, a UK startup that has developed an approach to AI based on a decade of research into insect brains.
Opteran reverse-engineers biological systems onto silicon, enabling machines to see, sense, navigate, avoid collisions, and make decisions better than existing AI approaches. It calls this 'natural intelligence' — also known as brain biomimicry.
I spoke to David Rajan, Opteran CEO, to find out more.
"Nature has already solved autonomy."
The Opteran team argues that nature solved many of the complex problems facing autonomous systems thanks to billions of years of evolution.
"If you really want to see the soul of autonomy, look at a garden, don't look in Google's data centre. Nature has already solved autonomy," asserts Rajan.
Opteran spun out of Sheffield University, and the birth of the idea was the European Union's human brain project — a €1 billion scientific research project funded by the European Union that ran for ten years from 2013 to 2023. It was one of the largest brain science projects ever undertaken, involving over 500 researchers from more than 150 institutions across Europe.
Rajan recalled:
"James thought that it was one of the most, forgive me, one of the most stupid ideas you'd ever heard of.
He thought if you're going to map a brain, don't pick one of the most complicated brains to start. Human brains are 86 billion neurons.
Pick something simple and learn how to do it. So James thought, "What would I do? I'd pick an insect like a honeybee with a million versus 86 billion neurons.
A honeybee can memorise and navigate to and from a destination, avoid obstacles and communicate journeys to other bees. o
It can see, like a human, perceive the world, moves around, and navigate using a pair of eyes, a tiny computer package that can use microwatts of energy."
Natural Intelligence offers a third wave of AI, challenging the prevailing belief that we can overcome all autonomy challenges with more data and computing power.
Opteran reverse engineers the natural brain algorithms that have evolved in biology over the last hundred million years, rather than gathering petabytes of data into thousands of servers and doing pattern matching using foundational models.
Rajan scoffs at the approach of industry heavyweights like NVIDIA, asserting:
"We're a little British company. We have one system and one software product with algorithms.
We reverse engineer from the brain. As a company, we don't have a data centre, and we don't even have an account.
And we reverse engineer algorithms from biology, and we put it on any machine."
The company's tech works on robotic dogs, drones, and ground-based robots "all running around with a general-purpose set of algorithms stolen from nature."
When Opteran's insect algorithms are implemented in a machine, the machine has an innate ability to move around successfully in a highly variable, dynamic environment.
Gaining commercial traction solving warehouse autonomous robot pain points
Importantly, the company is not stuck at the pain point of many university spinouts, struggling to gain commercial traction. The company recently announced that German autonomous picking and transportation robot manufacturer mobile robot (AMR) manufacturer SAFELOG will integrate Opteran Mind, the company's general-purpose autonomy product, into its AMRs.
Autonomous mobile robots (AMRs) operate in warehouses where existing autonomy systems struggle to distinguish between pallets and featureless and aliased spaces, especially when the light changes dynamically between areas such as loading bays and warehousing shelving.
Current solutions using 2D and 3D LIDAR, as well as vSLAM ( visual Simultaneous Localization and Mapping), are prone to location errors.
This is exacerbated when hundreds of robots operate together in a warehouse setting. Each installation requires an infrastructure consisting of magnetic tracks and QR codes or reflectors, increasing commissioning time and operating costs.
Opteran’s edge-only software uses low-cost cameras and silicon. Significantly, Opteran estimates its solution could cost less than $160 running on a Sony and ARM Core using Sony IMX219 cameras and RK4566 ARM chips. This contrasts with current state-of-the-art solutions, which can range in cost from $8,400 for a 2D LiDAR setup 1 to $27,000 USD for a 3D LiDAR setupsetup.
Opteran's localisation software enables new projects to be activated quickly and efficiently without additional infrastructure.
Rajan explained how the partnership came about:
"SafeLog's management directors are quite savvy. They understand the world, and they're not particularly sold on the AI narrative.
I think we caught their interest by saying, look, we're the anti-AI. We're doing it a different way. Little things like that can help you catch somebody's ear.
The rubber hit the road when we took our technology to them, and they couldn't break it.
You could lift the robot off the ground, move it around the room, and put it back on the map somewhere else, and it would figure out where it was in a second and carry on.
They were shining the light from their torches into the cameras, which was still working.
In the end, they covered the lenses with their hands to prove to themselves that we were using visual navigation."!
The company's tech is relevant to various industries, including drones used in aerial inspection, oil, and gas, where it can be hard to achieve robotic autonomy in dirty, low-light, dangerous, and challenging environments.
Perhaps its biggest claim to fame is that it's also captured the interest of the automotive sector.
Rajan explained:
"We are starting to do specific things with companies in that field, including advanced driver assistance and autonomous vehicles.
They are responding very well to the notion of using low-cost cameras or even the existing cameras.
They've already got a new car to do near-field vision and localisation. So, the ability to replace near-field Lidar and ultrasonics, which they have a lot of and cost a lot of money, is compelling. Many use cases exist around near-field vision, and we eliminate a vast cost barrier.
The big one for them is being able to do things like autonomous valet parking in an underground garage where there's no GPS signal."
Next on the horizon for Opteran are higher-order functions such as decision making. Rajan detailed:
"We've done natural decision-making in the lab. It allows machines to choose and prioritise what they want to do. So we want to roll that out.
Everybody's trying to get this moving around stuff right. The next thing is, when you're moving around, you've got to make some decisions.
How do you do that without a massive data set to support a reinforcement learning model?
How do you have a natural brain algorithm that takes input from the world and allows you to choose what to do?"
The company also aims to do more working on learning, image recognition, and abstract concepts like gravity.
"Nature is innately adaptable. It's about effectively solving the adaptability problem."
Would you like to write the first comment?
Login to post comments