Cerebionics wants to turn human intent into commands for machines

Sometimes you hear about a startup doing something just so completely out of left field that you have to find out more. For me, that startup is Cerebionics.

What started as Norwegian founder Agnessa Pedersen building drawing robots as a teenager has evolved into a brain-computer interface (BCI) platform she has spent months testing and refining with users in Ukraine.

I met with Pedersen at the recent TechBBQ conference to learn more.

I’ve always been interested in BCI. I’ve visited automotive companies using BCI to test driver response times for new software. I also once interviewed a designer working on an idea to enable people with quadriplegia to operate vehicles using BCI — from turning on the radio to more functional tasks and eventually steering the car. I had many emails from people with quadriplegia telling me how much they missed driving.

Crucially, non-invasive BCI doesn't need to reproduce fine motor control to be useful — it only needs to reliably capture intent, and Pedersen is taking that proposition out of the lab and into demanding real-world environments.

From drawing robots to brain-computer interfaces

Pedersen describes her foray into BCI as “super random.”

In middle school and high school, she became deeply interested in robotics and took online courses on platforms such as Udemy to learn Python and C++. She then began building her own robotic arms, initially because she wanted to create robots that could draw for her.

“I’ve always really liked drawing, especially big drawings, but they take a really long time. I thought it would be a neat idea to have an armada of robot arms at home that could lay down the first layers while I was at school, and then I could come home and take over.”

From there, she became interested in human-machine interfaces and how humans can remove the barrier between humans and machines. The idea of cyborgs got her interested in whether there was a way to extract your intent from your mind- what you want to do- and translate that into another physical system.

Pedersen started with wearables — specifically flex-sensor gloves — but they didn’t meet her goal of transferring intent to an end system, “because intent comes slightly before execution. I wanted to get closer to that.

“I wanted to connect our minds to machines and remove the need for physical control.”

Pedersen studied an integrated Master’s in engineering, specialising in cybernetics and robotics. Her studies combined control systems theory with programming, electrical engineering, and mechanical engineering, focusing on how these disciplines work together. During this time, she became interested in brain-computer interfaces.

“I started in 2024. I was still a full-time student then so I couldn't go all in.

“Then, about a year ago, when Project Europe came in, I decided to go full-time. I moved to London and built the team.”

“We’re building a highway between the brain and the end system”

Cerebionics is developing a modular, non-invasive BCI platform designed to connect with a wide range of systems.

“It doesn’t matter whether it’s a drone, an unmanned ground vehicle, or a C2 API. We’re essentially creating a highway between the brain and the end system, but doing it non-invasively,” explained Pedersen.

The technology uses EEG electrodes and is designed to be portable and lightweight enough to use while moving around.

“The advantage of EEG is that it’s a very stable modality that has been used extensively for a long time. BCI is difficult, but researchers have studied it for decades. In a way, we’re taking established technology and applying it to a new purpose rather than having to create everything from scratch,” she shared.

Different sectors, including defence, healthcare, consumer applications, and industry, all possess different usecases.

Cerebionics isn't trying to be Neuralink

While most people think of Neuralink when it comes to BCI, Cerebionics is pursuing a fundamentally different technical approach and set of use cases. Neuralink focuses on patients with severe conditions such as locked-in syndrome and quadriplegia, using thousands of electrodes implanted directly in brain tissue to capture neural activity at much higher resolution. That allows it to decode things like fine motor movements or, in some cases, aspects of speech and language.

Cerebionics uses non-invasive EEG, so it doesn't aim to achieve that same level of resolution. Instead, it detects broader patterns of brain activity associated with an intended action and translates them into higher-level commands for a machine.

“For our use cases, we don't need to decode the precise movement of an individual finger; we need to identify the user's intended command and pass that to the system.”

Cerebionics can do high-level commands such as movement in different directions, and it's working on expanding that dictionary, slowly building out the number of commands. Further, with non-invasive BCI, the signal quality is lower than it would be if it were implanted.

“That's simply a fact because of the skull and the way the technology works today,” explained Pedersen.

Why control machines with your brain?

You might be wondering why BCI is needed when muscle sensors, voice commands, and eye tracking already offer ways to control systems.

“There are cases where they're very useful, but I feel BCI is the final interface that is still relatively untapped. It opens up an additional layer of control that isn't being fully utilised,” shared Pedersen.

“I think it's good to have a collection of different command channels because they all have their strengths and weaknesses.”

One advantage of BCI is its potential to communicate intent directly. Rather than verbally describing exactly what an end system should do, an operator could transmit commands such as yes, no, confirm, dismiss, change direction, switch systems, execute, or abort. Cerebionics does this by capturing electrical activity in the brain and translating those signals into commands.

“Everyone has a pattern, although it changes somewhat from day to day. That's why we talk about calibration: currently, you need to calibrate the BCI because your pattern might be slightly different from one day to the next."

The more scientific answer: 

"We look at event-related desynchronisation and which areas in the brain activate.

The cool thing about the brain is that the signals you get from doing an action versus thinking about doing that action are very similar. That's part of why it's possible to have BCIs in the first place.”

The challenge is distinguishing deliberate intent from the brain's constant background activity.

“We've trained our models so we're able to decipher between noise and an actual, genuine intent command,” Pedersen said.

“Of course, we also have to implement safety layers. If the confidence is very low, then don't do anything. If the system isn't sure that it's a genuine command, it shouldn't act.

That also gets better the more an operator uses it.”

However, the latency between intent and action doesn't depend on the BCI alone. It spans the entire chain: acquiring the brain signal, decoding and classifying the command, transmitting it, and the end system receiving and executing it.

In an operational environment affected by jamming, for example, the BCI could function normally while communications with a drone or other end system are disrupted. Cerebionics can optimise signal acquisition and decoding, but overall response time also depends on the communications network, protocol, and end system. 

Making BCI work for everyone

For Cerebionics, the hardware is “good enough right now. We can get signals that we can work with.”

"The hardest part is the software: the decoder, and making the platform generalisable and transferable."

Ultimately, for this to scale into a mass market, the BCI platform has to be built so anyone can use it with minimal calibration.

“It can't cost $100,000 or $20,000 for the whole system and then require you to record huge amounts of data,” explained Pedersen.

For it to be adopted on a large scale, she predicts the overall price has to get down towards the price of conventional controllers, which vary substantially in price.

“That's tough, but not impossible.”

For BCI technology to scale, it needs to be generalisable and transferable across users and use cases For mass deployment, it also needs to perform as well as, or better than, existing interfaces while remaining easy to use. Setup and calibration need to be fast; the system can't be overly complicated, and users shouldn't need hundreds of hours of training to use it effectively.

On the degree of difficulty in training people to use the platform, when I spoke to Pedersen, Cerebionics hadn’t finalised a generalisable, transferable model yet.

“With that model, we expect onboarding and calibration to be very quick.”

At the time of the interview, it could take around an hour in an optimistic case where somebody understands it quickly.

“I've done BCI controls quite a lot, so I'm used to thinking in a way that suits the platform's functionality. But if you've never tried BCI before, it might be hard to wrap your head around at first. “We're looking at your intent, so learning how to do that can take some training. But that's true of any interface,” explained Pedersen.

She envisions the tech being offered in the future both as a standalone wearable, such as some form of headwear, and added into things people already use, such as smart glasses or goggles.

Europe’s emerging BCI startups

Europe is home to a small but growing group of startups developing different approaches to brain-computer interfaces. UK-based Cogitat develops device-agnostic AI software that decodes EEG brain signals into digital commands, with applications including robotic rehabilitation. Latvia’s BirgerMind is developing a non-invasive BCI designed to enable people with severe motor impairments, including ALS/MND and paralysis, to communicate and control digital devices using brain signals. Switzerland’s Neurosoft Bioelectronics is taking a minimally invasive approach by developing soft, stretchable electrodes designed to provide broad access to the cortex without penetrating brain tissue.

Sweden’s InnoBrain combines EEG, eye tracking, and AI to interpret cognitive and emotional states for applications ranging from human-machine interaction to operator monitoring.

Beyond EEG: the rise of non-invasive neurotech

The wider neurotechnology sector is also growing in scope, expanding beyond BCIs into technologies such as functional near-infrared spectroscopy (fNIRS) and focused ultrasound. fNIRS is a non-invasive brain-imaging technique that uses near-infrared light to measure changes in blood oxygenation in the brain.

Because active areas of the brain require more oxygen, these changes can be used as an indirect measure of neural activity. European companies, including Switzerland’s Optohive and the UK’s Gowerlabs, are developing wearable fNIRS systems that aim to make brain imaging more portable and usable outside traditional laboratory environments.  In Sweden, Mendi sits at the consumer end of the neurotech spectrum, using fNIRS to measure changes in blood flow in the prefrontal cortex and provide real-time neurofeedback through its brain-training platform.

Focused ultrasound takes a different approach. It uses precisely targeted sound waves to reach specific brain regions without surgery. Depending on how it is applied, the technology can modulate neural activity or temporarily open the blood-brain barrier, potentially allowing drugs to reach parts of the brain that are otherwise difficult to access. UK companies Mendea and NeuroHarmonics are exploring focused ultrasound for non-invasive neuromodulation, while France’s TheraSonic is developing the technology to improve drug delivery for neurological diseases. ​

“You can't build something for defence and only sit inside a lab in London”

Critically, Cerebionics isn't developing its technology solely in the lab. For Pedersen, understanding whether BCI can work in demanding operational environments has meant putting it in front of the people who might actually use it.

The team has spent months working with Ukrainian soldiers across different areas of the military, although understandably, much of the detail around that work has to remain off the record.

For Pedersen, Ukraine offers an opportunity to test Cerebionics in an environment where operational needs are immediate and constantly changing:

“In Ukraine, modern warfare changes every single day. The way the war looks now is completely different from two years ago.”

One encounter illustrates the potential use case particularly clearly. Pedersen recalled meeting a Ukrainian veteran who had been an FPV drone operator. During a combat mission, he lost most of his right hand.

“To fly an FPV drone in acro mode, ideally you need both hands. You have to control the throttle, pitch, and roll at the same time. He told me he felt that he'd lost his purpose because he'd been really good at it.

He really wanted to contribute to the fight against the adversary, but after his injury, he felt he could no longer contribute in a meaningful way.

For veterans, the transition back into civilian life can be extremely difficult.”

This scenario shows why an additional command channel could be useful. Rather than replacing conventional controls altogether, BCI could potentially provide another way for an operator to interact with a system when physical control is limited or other interfaces are already occupied. Pedersen sees defence applications more broadly as a potential way to enable operators to interact with multiple systems.

“When it comes to national security and defence, you have to constantly stay on the edge of your seat and build new capabilities. BCI is one of them.

BCI can enable operators to work more seamlessly with different systems, and it could, in a way, become a force multiplier. But that's more of a long-term goal.”

For Pedersen, developing those capabilities requires getting out of the lab and testing the technology in the environments where it might eventually be used.

“You can't build something for defence and only sit inside a lab in London. Those environments are so different.

Even within Ukraine, the eastern frontline is vastly different from the southern front. You might think it's the same country, so how different can it be? But there are a lot of differences."

Part of Cerebionics' competitive edge is that it spends so much time with end users. Pedersen said Ukrainian military personnel have been curious about the technology and willing to provide feedback, even when Cerebionics was at a much earlier stage.

“I'm actually surprised that nobody has ever told me to go away with my deep tech and said they don't need it.

When I first started going to Ukraine and talking to soldiers, what we were building wasn't perfect. We hadn't even thought about so many things the technology needed to be robust against.

I really appreciate their openness to sharing their experiences and what they've learned.”

Her approach to finding test users was equally direct.

“I went to events, found soldiers, and asked them.

People talk about needing to spend time with customers, but you have to get as close as possible to the ground truth. You have to see for yourself what it's actually like and put yourself in their shoes.

That's particularly true in defence and Ukraine. These soldiers are fighting for the sovereignty of their country. They're not going to sit on Zoom calls all day talking to startups.

Just go to them and talk to them.”

Testing has provided feedback on the user experience, performance metrics, and insights into what needs to change to make the technology more suitable for different operational situations.

“All of that has made us better.”

What happens to your brain data?

But putting BCI into real-world environments raises another question that becomes increasingly important as the technology moves beyond the lab: what happens to the neural data it collects? In Europe, there is not yet a dedicated regulatory framework specifically governing BCI data, although existing rules, including GDPR, can apply

Any neural data Cerebionics collects is anonymised, with location data and other potentially identifying metadata removed, and stored locally rather than in the cloud.

From the battlefield to gaming

Cerebionics has received commercial interest, with several letters of intent, but its current focus is scaling the team of four and pursuing long-term collaboration projects. But in terms of use cases, the sky’s the limit, especially where eye tracking, hand control, and voice are already engaged.

“Oh my God, gaming. That would be so cool,” enthused Pedersen.

“Imagine going to a gamer — say, a Call of Duty fan — and saying, ' This technology has been tested in a genuinely demanding environment; now have a go at it.”