When disaster strikes, mechanical snakes are often the ultimate rescue tool. Their slender, highly articulated bodies allow them to effortlessly navigate collapsed buildings, squeeze through narrow crevices, and traverse uneven rubble to locate survivors.
However, this incredible agility comes at a steep cost. Mimicking the undulating, slithering motion of a real snake requires multiple mechanical motors to coordinate simultaneously, rapidly draining the robot’s battery and severely limiting how long it can search a disaster zone.
Now, researchers in Japan have solved this critical flaw. Using deep reinforcement learning, scientists have trained an artificial intelligence to teach robots an entirely new, highly energy-efficient way to move.
Re-inventing the wheel
According to a new study from Osaka Metropolitan University, the AI discovered that the robots could dramatically conserve energy by abandoning their slithering motion entirely when on flat surfaces.
Instead, the AI commands the snakebot’s head and tail to connect, reshaping the robot into a complete circular structure. By subtly shifting its centre of gravity, the robot can rapidly roll across level ground, using gravity to do the heavy lifting rather than relying on constant, battery-draining motor power.
To ensure the robot does not veer off course, the team developed a unique “observation buffer.” This system constantly analyses the robot’s angular velocity, acceleration, and bodily state to stabilise the rolling motion and guarantee precise, straight-line travel.
Dr Akio Yamano, who led the research group at the university’s Graduate School of Engineering, explained that the results were striking.
“We found that on level ground, the rolling motion achieved approximately twice the travel speed per unit of power consumption compared with the undulating motion,” Dr Yamano said.
The AI smart mix
Because rolling is highly inefficient on jagged rubble, the researchers concluded that the ultimate strategy for extending battery life is a hybrid approach: slithering over uneven terrain and dynamically switching to a rolling motion the moment the robot hits a flat surface.
This mixed-movement strategy could drastically extend mission times for search-and-rescue operations in earthquake-prone areas, and could even be applied to future planetary exploration missions.
Looking ahead, Dr Yamano hopes to design autonomous robots capable of actively reading their environment to determine their optimal movement strategy in real time, rather than relying on a human operator to switch between pre-programmed gaits.
“Our group is developing various interesting capabilities,” Dr Yamano concluded. “We aim to create robots that autonomously assess the situation and use precise navigation technologies to carry out useful tasks.”
