Cloudy and turbulent waters often pose a challenge for remotely operated underwater vehicles. These vehicles stir up sediment when they settle on the seafloor or dig into sandbeds, obstructing the view of onboard cameras. Typically, operations have to pause until the sediment clears. However, a new mapping technique by MIT and the Woods Hole Oceanographic Institution (WHOI) engineers could enable vehicles to see through murky waters.<\/p>\n
This technique combines visual images from optical cameras with sonar sensor data. By doing so, vehicles can quickly map their surroundings using sonar, even in low-visibility conditions. They can then approach specific shapes within the sonar-mapped environment to capture detailed images with optical cameras. This method mirrors how dolphins use echolocation and sea turtles rely on close-range vision to navigate murky waters in real-time.<\/p>\n
The researchers tested the technique in controlled tank experiments, demonstrating its ability to visualize objects and map environments even in the cloudiest conditions. Known as Sonar-MASt3R, the method could guide underwater vehicles in murky conditions for scientific exploration, construction, maintenance, and recovery operations. Amy Phung, a graduate student at MIT, highlighted the potential for operations in challenging environments.<\/p>\n
Phung presented a paper on Sonar-MASt3R at the IEEE International Conference on Robotics and Automation (ICRA). Richard Camilli, a senior scientist at WHOI, co-authored the paper. Traditionally, underwater navigation relied on either optical cameras or sonar sensors. Optical cameras provide detailed images in clear water, while sonar sensors work well in murky conditions by identifying the shape and distance of objects through acoustic waves.<\/p>\n
Recent efforts have focused on combining these methods, known as “opti-acoustic fusion,” for object recognition and workplace mapping. Most existing techniques require time to sync and process data and do not operate in real-time or in 3D. Phung and Camilli aimed to develop a real-time, detailed 3D mapping technique for low-visibility underwater conditions, inspired by challenges such as safely recovering unexploded underwater mines.<\/p>\n
Sonar-MASt3R builds on a French-developed technique called MASt3R, an image-matching algorithm that estimates relative pixel depth to create 3D maps. Sonar data provides absolute scale measurements, allowing for precise 3D maps. In murky water, sonar-corrected maps enable vehicles to determine object locations and approach safely with optical cameras.<\/p>\n
Experiments involved a tank filled with water, sediment, and objects, using a robotic arm equipped with a camera and sonar sensor. An initial sweep with the arm collected data to create a rough map, which was refined with close-up images. The keyframe approach quickly integrated new visual details into the map. Sonar-MASt3R generated accurate 3D maps, resolving centimeter-scale details even in the cloudiest conditions where cameras alone failed.<\/p>\n
Camilli likened the process to navigating a dark china shop without causing damage. The team plans to test Sonar-MASt3R in natural underwater settings, anticipating that mapping will be more straightforward due to fewer distortions than in a tank. This technology could enable safer exploration of murky underwater regions.<\/p>\n