What does it mean for a robot to be “aware”? For animals, awareness is woven into every movement: sensing the environment, adjusting to changes, recognising when something is wrong. In artificial systems, those abilities don’t come naturally—they must be designed and measured. A new study, co-authored by Mi3 researcher Nadine Meertens together with colleagues at the University of Bristol, Suet Lee, Emma Milner, and Sabine Hauert, offers a practical way to do precisely that.
Rather than tackling the philosophical question of consciousness, the authors take a functional view. A robot is “aware,” in their framework, when it can gather information, interpret it, and use it to act. They break this into several dimensions—including spatial awareness, bodily awareness, temporal awareness, agentive awareness, and metacognitive awareness—and then propose a method for testing how adding abilities along a dimension affects performance in real tasks.
To demonstrate their approach, they turn to swarm robotics: ten small robots operating in a 5-by-5 metre arena, searching for boxes and teaming up to transport them to a drop-off zone. This intralogistics scenario is perfect for studying awareness, because the robots must navigate, collaborate, and cope with occasional faults—just as they would have to in a real setting.
The experiment focuses on two awareness types. The first is spatial awareness, tested in a simple way: robots either know only the general quadrant where a box is located or they know the precise coordinates. The second is bodily awareness. Here, robots can either detect that they are faulty without knowing why or, at a higher level, diagnose the exact fault—such as a broken wheel or malfunctioning lifter—and respond appropriately.
Over hundreds of simulated trials, the researchers measured how quickly each version of the swarm delivered boxes. Some findings were unsurprising: robots with precise spatial information worked much faster, and robots that could diagnose their own faults outperformed those with no self-awareness. But one result stood out. Robots that only knew “something is wrong with me,” without knowing what, did worse than robots with no fault detection at all. This partial awareness often led them to choose the wrong corrective action, slowing or even blocking the team. In other words, a little awareness can be more harmful than none.
The study, part of the EU-funded EMERGE project, also revealed that different types of awareness can enhance each other. When robots combined partial spatial information with full fault diagnosis, the overall performance exceeded what either ability provided on its own—a sign that awareness dimensions can interact in useful, non-obvious ways.
Although the paper uses a robot swarm as its test bed, the framework is much broader. It provides a way to talk about, measure, and compare awareness across many kinds of artificial systems, from soft robots with tactile sensors to neural networks making predictions over time. Ultimately, it suggests that engineers might one day design robots not just with specific behaviours, but with specific awareness profiles, a structured set of abilities tailored to the job at hand.
The work shows that awareness can be a practical design element with measurable consequences. And by treating awareness as something engineers can tune and test, the authors open the door to building artificial systems that are not only more capable, but more adaptive, reliable, and responsive to the world around them.
Check out the original article and Nadine Meertens’ profile.
Reference:
Lee, S., Meertens, N., Milner, E., Hauert, S. (2026). A Framework for the Examination of Awareness in Artificial Systems. In: Jiménez Rodríguez, A., et al. Biomimetic and Biohybrid Systems. Living Machines 2025. Lecture Notes in Computer Science, vol 15582. Springer, Cham.