Researchers from Tohoku University in Japan and the Vidyasirimedhi Institute of Science and Technology (VISTEC) in Thailand have developed a method that allows a six-legged robot to learn locomotion from stick insect walking data. The robot learned to walk within about an hour and was also able to traverse uneven terrain and adapt to the loss of a limb.
The researchers used an open dataset containing three or four steps recorded from a stick insect. Rather than directly reproducing the insect’s movements, an artificial intelligence system used the data to infer the objective behind the animal’s walking and determine how leg movements could achieve that objective.
The approach is based on inverse reinforcement learning, in which an AI system attempts to infer the reward or objective underlying an example of desired behaviour. This differs from methods in which developers explicitly define how a robot should move its individual legs.
“We never told the robot how to walk,” said Dai Owaki, associate professor at Tohoku University. “We asked what the insect was trying to achieve, and let the robot chase the same thing entirely on its own.”
The researchers applied the resulting method to RedMirror, a six-legged robot developed at VISTEC. According to the research team, the robot learned to walk three times faster than when trained using a standard reward.
The system separates what the researchers describe as body-independent information from information specific to an individual machine. This is intended to allow the learned locomotion principles to be transferred between robots with different physical characteristics instead of requiring a new walking strategy to be designed for each platform.
The experiments also tested the robot under conditions beyond normal walking. RedMirror was able to negotiate uneven terrain and modify its locomotion when one of its legs was unavailable, indicating that the learned strategy could accommodate changes in the robot’s body and surroundings.
“It’s remarkable that a few steps from a single stick insect were enough to find a principle that works on a machine five times its size,” Owaki said.
The researchers plan to add memory capabilities so robots can accumulate experience over time. They see adaptive legged locomotion as potentially applicable to robots operating in environments such as disaster sites, where uneven terrain can restrict the use of wheeled systems.
Photo credit Dai Owaki, 2026
