Our attempts to understand the ocean still need to be more superficial. According to National Oceanic and Atmospheric Organization, around 80 percent is “unmapped” and “unobserved.”
The primary method of collecting ocean information is sending ships, but it’s expensive to do so frequently. Argo floats, a new generation of robotic buoys, have drifted with the currents and taken measurements up to 6,500 feet deep. New aquatic robots developed by a Caltech lab could go deeper and perform more specific underwater missions.
John O. Dabiri is a professor of aeronautics, mechanical engineering, and computer science at the California Institute of Technology. He says, “We are imagining a global ocean exploration approach where we take swarms of small robots of different types and populate them with the ocean for tracking, climate change, and understanding the physics of the ocean.”
CARL-Bot is a palm-sized aquatic robotic resembling a cross between a dumbo Octopus and a pill capsule. It has motors to swim around and weights that keep it upright. It also has sensors for detecting pressure, depth, and acceleration. CARL’s microcontroller, with a processor of 1 megabyte and a size smaller than a stamp, is the engine behind everything it does.
Peter Gunnarson, a Caltech graduate student, created and 3D printed CARL at home. Gunnarson’s first tests were done in his bathtub, as Caltech labs will be closed by COVID at the beginning of 2021.
CARL is still remotely controllable. To reach the deepest parts, the hand-holding must stop. This means that researchers cannot give CARL instructions. It must learn how to navigate this vast ocean by itself. Gunnarson, Dabiri, and Petros Koumoutsakos worked together to develop AI algorithms that would teach CARL to navigate using its environment and previous experiences. The research was published in Nature Communications this week.
CARL may change its route to avoid rough currents and reach its destination. It can also stay in place at a specific location using “minimal” energy from a battery.
The power of CARL lies in its memories.
Koumoutsakos’ algorithms can do the calculations for the robot. The algorithms take advantage of the robot’s memory, which can be used to navigate past a whirlpool or other obstacles. Dabiri explains, “we can use this information to decide how to navigate these situations in the future.”
Gunnarson says that CARL can remember the paths it took in previous missions and, with repeated experience, “get better and better at sampling the oceans using less time and energy.”
In simulations, all data points are cleaned. Transferring machine learning to the real world can be messy. Sensors can be overwhelmed and may only capture some of the metrics. Gunnarson says, “We are just beginning the trials in the tank.” First, CARL will be tested to see if it can perform simple tasks like diving repeatedly. The robot is shown in a video posted on Caltech’s blog bobbing and diving into a tank of still water.
The team will test CARL in a tank-like pool with jets to create horizontal currents. The robot will then graduate to a two-story facility that mimics upwelling and downward currents. It will need to find a way to maintain a specific depth in an area where water flows in all directions.
“Ultimately, however, we want CARL to be in the real world. Dabiri says that he will leave the nest to go to the ocean, and after repeated trials in the sea, the goal is for him to be able to navigate by himself.
During testing, the team also will adjust the sensors on and in CARL. Dabiri says, “One of our questions was what’s the minimum set of sensors you can have onboard to complete the task.” A robot with tools such as LiDAR and cameras can’t stay in the ocean for long before the battery needs to be changed.
Researchers could extend the life of CARL by reducing its sensor load. They would also have more space for scientific instruments that measure pH, temperature, salinity, and other parameters.
CARL’s software may inspire the next generation of bionic jellyfish
Dabiri’s group published an article early last year on how they used electrical zaps to control the movements of a jellyfish. Researchers could improve the ability to steer jellies by adding a similar chip to CARL.
Dabiri says it could take time and effort to figure out how the navigation algorithm functions on a live jellyfish. CARL is a vessel that can be used to test the algorithms which will eventually go into mechanically modified creatures. These jellies would not be limited by depth, unlike robots or rovers. Biologists already know these creatures can survive in the Mariana Trench – a depth of 30,000 feet.
CARL can be a valuable asset for ocean monitoring. It can be used with existing instruments, such as Argo floats. Or it can go on solo missions for more detailed explorations. It can track and tag along biological organisms such as a school or fish.
Dabiri says, “Imagine in the future 10,000 or a billion CARLs (we will give them other names) going out to the ocean simultaneously to measure areas that we can’t reach today to get a more time-resolved view of the ocean’s changing.” This will be essential for climate predictions and understanding how the ocean functions.
