Engineers from the California Institute of Technology (CalTech) have developed an AI algorithm to help an autonomous drone swim, using ocean currents to aid its navigation. Eventually, these marine robots could explore the oceans of Earth and other worlds, such as Europe, by monitoring the conditions of environments inaccessible to humans.
“When we want robots to explore the deep ocean, especially in swarms, it’s almost impossible to control them with a joystick 20,000 feet from the surface,” said aeronautics and engineering expert Professor John Dabiri. mechanics at CalTech. “We also cannot provide them with data on local ocean currents that they need to navigate because we cannot detect them from the surface.
“Instead, at some point we need ocean drones to be able to make decisions about how to move on their own.”
These drones should be able to decide for themselves where to go, but also the most efficient way to get there. To do this, they will probably only have data that they can collect themselves: information about the water currents they are currently experiencing.
The engineers therefore developed an algorithm that would allow the drone to steer while swimming. For this, they used reinforcement networks. Compared to conventional neural networks, reinforcement learning networks do not train on a static data set, but rather train as fast as they can gain experience.
This scheme allows them to exist on much smaller computers. To test their algorithm, the team wrote software that can be installed and run on a “Teensy” (a microcontroller anyone can buy for less than $30 on Amazon that runs on half a -watt). These types of microcontrollers can fit on a palm-sized prototype robot – which they used alongside computer simulations to test their work – mimicking the hardware of possible future maritime drones.
Using a computer simulation in which passing an obstacle in the water created multiple vortices moving in opposite directions, the team trained the AI to navigate in ways that took advantage of regions at low speed in the wake of the vortices to steer towards the target location with minimal power used. To aid his navigation, the simulated swimmer only had access to information about the water currents at his immediate location, but he quickly learned to exploit the eddies to steer towards the desired target. In a physical robot, the AI would similarly only have access to information that could be gathered from an on-board gyroscope and accelerometer, which are both relatively small and inexpensive sensors for a platform. – robotic form.
This type of navigation is, engineers say, analogous to the way eagles and hawks ride thermals in the air, extracting energy from air currents to maneuver to a desired location with the minimum amount of energy expended. Surprisingly, the researchers found that their reinforcement learning algorithm could learn even more efficient navigational strategies than those thought to be used by real fish in the ocean.
“Initially, we just hoped that the AI could compete with the navigational strategies already found in real swimming animals, so we were surprised to see it learn even more efficient methods by exploiting repeated trials on the computer,” said Dabiri said.
The technology is still in its infancy; Currently, the team would like to test the AI on every type of flow disturbance it might possibly encounter during an ocean mission – for example, swirling eddies versus continuous tidal currents – to assess its effectiveness in nature. However, by incorporating their knowledge of ocean flow physics into the reinforcement learning strategy, the researchers aim to overcome this limitation.
“Not only will the robot learn, but we will learn about ocean currents and how to navigate through them,” says Peter Gunnarson, graduate student at Caltech and lead author of the book. Nature Communication paper.
Sign up for the E&T News email to get great stories like this delivered to your inbox every day.