Researchers from Osaka University in Japan and Diponegoro University in Indonesia have developed an artificial intelligence system capable of helping cyborg cockroaches navigate complex terrain more efficiently. The project employs Madagascar hissing cockroaches equipped with a small electronic backpack containing sensors and an AI system.

The technology was created to address a limitation of traditional cyborg insect systems: typically, these animals are steered via electrical stimuli applied to their bodies—stimuli that can alter the insect's direction, allowing an electronic system to control some of its movements. However, continuously controlling an animal in this manner is not always effective when it encounters obstacles or uneven terrain.

The new approach allows the AI ​​itself to decide when to intervene in the cockroach's movements and when to let the insect's natural behavior take over; to achieve this, the backpack mounted on the animal uses sensors to analyze the terrain as the cockroach moves.

This information is processed by a multilayer perceptron neural network capable of classifying the environment into four distinct scenarios: flat ground, an incline, a decline, or a potential fall. In tests, the system successfully recognized these conditions with approximately 92% accuracy. Identifying the terrain allows for the automatic adjustment of how electrical stimuli are applied.

For instance, when the cockroach encounters an obstacle it needs to climb, the system lowers the intensity of the electronic control; this enables the animal to use its natural behavior to overcome the obstacle. In experiments, this combination reduced hesitation and unnecessary movements, allowing the insects to traverse complex terrain more quickly and efficiently.

The team describes this approach as a form of bio-hybrid physical AI—and yes, I know you thought these insects were being used for espionage or something similar; right now, as you think out loud, one can almost "hear" you imagining humans being controlled by AI.