On land and in the air, there are some very advanced methods available for the localization and navigation of autonomous robotic systems. But these technologies, which often rely on GPS signals or computer vision, are not very useful in underwater environments. Radio waves do not travel far enough through water to be of much help, and cameras face similar problems due to turbidity or low light levels. This leaves underwater robots high and dry when it comes to exploring the depths.
Given the physical limitations imposed by water, engineers often turn to acoustic sensing for navigation. In much the same way that dolphins and other aquatic animals find their way around, we can transmit sound waves through the water, then listen for the reflections to determine how far the transmitter is from nearby objects. However, the accuracy of these techniques is not very high, especially when compared with technologies like GPS and computer vision.
But these acoustic systems could improve in the future, thanks to an algorithm called Certifiably Optimal RA-SLAM (CORA) that was developed by researchers at Northeastern University. The algorithm not only determines how far the sensor is from nearby objects, but it uses that information to determine where in the water the robot actually is, which is necessary for localization. This is a capability missing from most existing techniques, rendering them much less useful for real-world applications.
An overview of the algorithm (📷: Northeastern University)
In particular, the algorithm dramatically improves the performance of a method known as range-aided simultaneous localization and mapping (RA-SLAM). This technique enables underwater robots to map their surroundings while keeping track of their own position using distance measurements from acoustic sensors. But while RA-SLAM has potential in underwater exploration, its usefulness has historically been limited by unreliable data and the mathematical difficulty of solving its core equations.
While traditional RA-SLAM solutions are sensitive to initial guesses and prone to getting stuck in suboptimal results, CORA side steps these issues through a new mathematical method. By transforming the problem into a quadratically constrained quadratic program and then relaxing it into a semidefinite program, the algorithm becomes far more robust. Using a method called the Riemannian Staircase, CORA can solve this relaxed problem efficiently and refine the results into a usable estimate of the robot’s location.
In addition to performance enhancements, this system also has the potential to make high-quality underwater navigation more affordable. While traditional systems can cost upwards of $500,000, the use of CORA with lower-cost acoustic sensors could reduce that to about $10,000.
CORA has been successfully tested on autonomous surface vehicles in the Charles River, proving that it could be a valuable tool for scientists in many fields, including those studying the climate. The team points out that the underwater robots used in the Arctic to study glacial melt could be given more reliable navigation systems with the help of CORA. Exactly where the technology goes from here is uncertain, but it has the potential to help robots reliably reach challenging environments to record previously inaccessible data.