Scientists from the Delft University of Technology in the Netherlands and Wageningen University have developed a system that enables autonomous vehicles to detect an approaching failure, much like how living organisms sense pain. The authors believe this approach could become a key tool for enhancing safety in both drones and self-driving cars.
The concept is rooted in a simple biological principle. When a person gets injured, their nervous system instantly signals danger, prompting them to reduce strain and prevent further damage. The researchers sought to replicate a similar mechanism for machines by creating a kind of digital counterpart to the pain system.
The findings were published in the journal Proceedings of the National Academy of Sciences (PNAS). The technology is based on a phenomenon known as “critical slowing down.” This method was previously used in ecology to forecast catastrophic changes in complex systems, such as the collapse of ecosystems. The scientists discovered that similar patterns emerge in technical systems shortly before they lose stability.
The main distinction of this new method is that it does not require pre-built system behavior models, historical data, or complex predictive algorithms. Instead, the technology continuously analyzes real-time signals from existing sensors and identifies characteristic changes that may indicate an impending failure.
To test the method, the researchers used quadcopters from the CyberZoo research facility. During experiments, they progressively damaged the propeller blades, increasing the level of destruction up to 55%. It was found that the drone began to lose controllability when approximately 15% of one of the front blades was damaged. Meanwhile, the developed system successfully detected signs of instability before the critical state was reached.
Study leader Jasper van Beers compared the algorithm’s operation to the human experience of pain. According to him, pain provides immediate feedback about the body’s condition and helps determine which actions remain safe. Modern machines, in contrast, largely lack this kind of self-diagnostic capability.
The authors believe that autonomous cars and driver assistance systems represent a particularly promising area for this technology. A self-driving vehicle might face gradual sensor failures, degradation of actuators, or complex road conditions, yet current systems often only detect the problem once it has already started to affect driving safety.
Since the new approach relies solely on data from standard sensors, its potential implementation would not require installing additional equipment. This makes the technology especially attractive for existing autonomous vehicle platforms, including robotaxis.
According to the researchers, endowing machines with a kind of “sense of pain” could be a significant step toward creating truly safe autonomous systems capable of recognizing their own limitations before a situation becomes critical.
About the Author
