Forget commands – this robot dog figures it out on its own

Swedish startup IntuiCell says it has “translated” biological intelligence into software that allows robots to learn like living beings.

Contrary to the familiar saying, an old dog can learn new tricks – especially if it’s equipped with cutting-edge software that adapts to the environment like a human brain or that of other animals.

Stockholm-based IntuiCell, a spin-off of Sweden’s Lund University, has created what it calls the world’s first functional digital nervous system, a “genuine” non-biological intelligence capable of real-time, real-world learning.

To show it off, it used an off-the-shelf robot dog from Unitree, nicknamed Luna. Without any pre-training or instructions, Luna taught itself how to stand for the very first time – much like a baby deer wobbling up on its legs.

What makes this special is that IntuiCell’s system doesn’t rely on static training data like traditional AI. Instead, it mimics a biological nervous system, meaning Luna learns through experience, trial and error, and direct interaction with its surroundings.

In demos, Luna adapted to tricky terrain, like rocks and even ice – without needing to retrain for each new task. The company says this could be a major step toward truly autonomous machines that learn in real-world “chaos,” not just clean lab environments.

Robot intuition?

According to Linus Mårtensson, lead developer and co-founder of IntuiCell, a traditional language model works by collecting vast amounts of data – essentially everything humans have ever typed on the internet – and learning to predict what comes next.

“It’s learning statistical patterns of how others think and create text and it's copying those patterns. It’s not learning by itself to do new things, but learning to replicate what others already do,” Mårtensson says.

In contrast, IntuiCell’s invention is not a language model but a system designed to imitate biological learning, something that could be called intuition.

“We’ve built a system that more closely looks like what happens in biology and learns from its inputs – directly from its experiences and actions in the real world,” Mårtensson tells Cybernews.

“We’re basing our whole learning framework around being able to learn in the real world, instead of learning from what others have done,” he says.

From chores to Mars exploration

The technology behind IntuiCell’s robot dog isn’t limited to quadrupeds. According to the company, the learning system is “so generic” that it can be applied to any agent – physical or digital – including drones and humanoid robots.

Its ability to learn on the spot and adapt to new environments opens the door to a wide range of applications, from household helpers to Mars missions.

“Most likely, you don’t want the exact same assistant in every household,” Mårtensson says, noting that everyone has different needs – and a robot equipped with this technology will be able to adapt to them, not the other way around.

The adaptive nature of the system makes it especially promising in environments where pre-programmed responses and static training data fall short. “In the long term, I would really love to see this used in environments where it’s hard to collect data,” Mårtensson says.

“Say someone wants to send the first humanoid robot to Mars or into deep space – you can't have a system that's been trained on every potential outcome from the start. It has to be able to realize there's an unforeseen situation, adapt to that situation, and solve problems as they go.”

The same approach could be life-saving here on Earth.

“You could see it in households, you could see it in space, you could see it in disaster response,” Mårtensson says.

“In those environments, you're never going to be prepared for every eventuality. But if we can send in robots that adapt to unforeseen circumstances, then we save lives.”