Researchers in the Autonomous Systems and Controls Laboratory, including Abhilash Chowdhary, Stephen Krauss, and Scott Gibson, are developing underwater robots for search missions.

Whether a task uses our human senses or robotic ones, resources are always limited. When the task is at the bottom of the ocean, resources are even more limited. ECE Professor Dan Stilwell and his team are addressing fundamental questions in search theory that are leading to new ways to deploy perception resources. With collaborators from aerospace and ocean engineering, including Professor Wayne Neu, they are developing new classes of autonomous underwater vehicles (AUVs) that will eventually be used for efficient subsea search.

Eventually, their research will lead to teams of robots that will be able to autonomously carry out a search mission in unknown waters. According to Stilwell, the robots might be looking for natural resources, airplane parts, or even underwater mine fields.

"At this stage it's a math problem," says Stilwell. "We assume that perception sensors are error prone, and that we cannot search exhaustively," he continues, "so we need the autonomous vehicles to make the best decisions possible about where to search, and they need to make these decisions without human intervention." To make the problem more challenging, the number of items to be discovered is unknown, so it's hard to determine when the search is complete.

These search robots will need to communicate with each other, which is particularly challenging underwater. The team has integrated acoustic communication systems into the robots which allows them to communicate by sound. Navigation, which is the process of determining location, is also difficult for an underwater robot. "When a vehicle goes underwater, you don't have GPS, so you need to find other methods," explains Rami Jabari, a masters student in electrical engineering. The team's navigation solution uses measurements from a variety of sensors that can be available at different times during a mission. Sensors that aid navigation include inertial sensors that measure the robot's acceleration and angular velocity, a Doppler velocity log that measures the speed of the AUV from the Doppler shift of sound after it bounces off the seafloor, and an acoustic range-tracking system that use acoustic signals to measure the range to beacons placed on the seafloor or on nearby boats. The team is also developing ways to incorporate measurements of the underwater terrain to aid navigation.

Part of the challenge of working with underwater robots is creating systems that can withstand such a harsh environment, like their current prototype vehicle pictured here.

Although the underwater robots can communicate with each other and with a human operator, subsea communication is very unreliable, and the team must develop autonomy algorithms that assume no communication with a human operator. "When we turn our systems on, they disappear on purpose. Our sense of risk tolerance is lower than other laboratories," Stilwell says. Underwater robots must function in a very harsh environment, he adds. This impacts communication, propulsion, and control.

Not only do the search robots face communications limitations, but they also have severe energy limitations, according to Jack Webster, an aerospace and ocean engineering Ph.D. student in the lab. Underwater robots are powered by batteries, which must be placed in watertight housings designed to withstand the immense pressure of the ocean. This limits the space that can be devoted to batteries, and it makes energy efficiency a critical design goal, especially for propulsion.

To address the challenge of efficient propulsion, Webster is investigating small propellers that operate at low speedsspeeds that have not previously been studied as thoroughly as the faster and larger propellers used in airplanes and ships. "I'm working on developing a computational fluid dynamics model of the propulsion system," says Webster. "We need to incorporate turbulence and similar effects, but small and slow propellers have been mostly ignored by researchers in the past, and there is significant uncertainty about what sort of model should work well for our application."

For the control system, electrical engineering doctoral student Scott Gibson is developing improved dynamical models of underwater robots and improved processes for acquiring them. Better dynamic models are needed to support the development of high-performance control systems, and Scott also hopes to significantly reduce the time and resources needed to develop a high-quality model. "The biggest challenge," according to Gibson, "is gathering data from vehicle maneuvers that unambiguously display the dynamics we are trying to identify. In the end, we hope to have new fundamental results on dynamical models for underwater vehicles, but also practical results on how to acquire and use the models."

Although much of this work is still at a fundamental level, the technology is progressing quickly, according to Stilwell, who expects some aspects of the technology to be ready for commercial deployment in a couple years. "We're close," he notes.