Monday, July 4

A robot of a span that will help us to know the oceans, nothing better than fish and aims at space exploration: this is CARL-Bot

It does not lift more than a foot and, apparently, it looks more like a kind of pill than a complex robot with Artificial Intelligence (AI), but CARL-Bot, the device that a group of researchers from Caltech, ETH Zurich and Harvard are working on, promises to revolutionize the study of the oceans. Who knows, it may even be that of other planets with wide expanses of water, such as Enceladus or Europa, satellites, respectively, of Saturn and Jupiter.

Unlike other robots and devices used by scientists to study the oceans, CAR-Bot offers a special peculiarity: through artificial intelligence its creators they want it to be autonomous and to be able to move freely by ocean currents. Not like a drifting beacon would, but by monitoring water conditions and taking advantage of reinforcement learning (RL) networks, which, unlike conventional neural networks, are not trained with a set of statistical data, but based on accumulating experience.

A unique ally for the depths of the ocean

It may seem like a new twist, but it is actually a fundamental skill for deepen knowledge of the oceans, especially in those regions that are more inhospitable, inaccessible and that have been least investigated to date. “When we want robots to explore the depths of the ocean, especially in swarms, it is almost impossible to control them with a joystick from 20,000 feet away. [alrededor de seis kilómetros] on the surface. Nor can we provide them with data on the local ocean currents they need to navigate because we cannot detect them from the surface, ”explains John O. Dabiri, Professor of Aeronautics and Mechanical Engineering, in an article published by Caltech University.

John Dabiri and Peter Gunnarson testing CARL-bot at Caltech.

What researchers need are, Dabiri emphasizes, drones trained to “Make decisions about how to move themselves”. And the CARL-Bot robot aims precisely at that ambitious goal. AI performance has already been tested using computer simulations that emulated vortexes and has taught you to take advantage of low-speed regions by assessing information from your immediate environment. The reason? In the future CARL-Bot will depend primarily on the data you collect with a gyroscope and accelerometer.

The system is similar to that used by eagles and hawks themselves to move between air currents. “Surprisingly, the researchers found that their reinforcement learning algorithm could learn navigation strategies that are even more effective than those believed to be used by real fish in the ocean.” the article published by Caltech itself: “Initially we just hoped that the AI ​​could compete with the navigation strategies already found in real swimming animals, so we were surprised to see that learned even more effective methods by exploring repeated essays on the computer ”.

The research, in any case, is still in development and must confront the AI ​​to the different scenarios that could be found in the open sea, such as swirling or flowing vortices. In their favor, the research team has the experience you have already gained from working with jellyfish. Initial results have been published in Nature Communications.

In addition to refining the AI, the team has started experimenting with the robot itself. After a first test in a small bathtub, a well-controlled environment, it has moved to a larger tank and in the future they plan to test the device in a pool with jets capable of generating horizontal currents and in which CARL-bot can move with vertical movements or learn to stay at a certain depth with various currents.

“Ultimately, we want CARL in the real world. It will leave the nest and go to the ocean and, with repeated tests there, the goal would be for it to learn to navigate on its own ”, explica Dabiri a Popular Science. To ensure it will be kept in a small footprint to make it easier to work or even use “swarms” of robots, the researchers developed software that can be installed on a 2.4 to 0.7-inch Teensy. The goal is for your movement to be as efficient as possible, but you can remain in a controlled environment thanks to a lithium ion battery.

For now, the result, which you can see on video, is surprising.

Cover image | Caltech

Leave a Reply

Your email address will not be published.