It’s no longer a surprise to see driverless cars bopping around the neighborhood. Fish, however, have a while to wait before autonomous robots enter their habitat.

When it comes to navigation through reinforcement learning — strategies learned through trial-and-error interactions with the physical environment — the leap from land to water is greater than you might think.

Land-based autonomous vehicles use onboard sensors to pick up global positioning signals (GPS). Underwater, however, the absence of GPS and the dynamic nature of currents mean that freewheeling robots and self-motivated submarines are still science fiction.

For the Kanso Bioinspired Motion Lab, based at the USC Department of Aerospace & Mechanical Engineering, technological problems can often be solved by learning from nature.

The lab specializes in the study of the flow physics of living systems. Researchers combine mathematical modeling with experimental data to identify how cells and organisms have evolved to interact with fluid environments — primarily air and water — revealing optimal survival strategies that can advance engineering solutions.

“Aquatic animals are much better at underwater navigation than robotic vehicles,” said Professor Eva Kanso, the Z.H. Kaprielian Fellow in Aerospace and Mechanical Engineering. “Our lab asked the question: Could a robot learn about its surroundings from its own trial-and-error interaction with a fluid medium — what we call an ‘egocentric’ approach — as opposed to receiving data from an external information source?”

Just imagine the possibilities if robots were able to swim as freely as fish. Stormy waters would pose no problem for search-and-rescue missions; underwater infrastructure could be fixed more efficiently; we could gain a richer picture of the seafloor and accurately track changes in biodiversity and climate.

The Kanso Lab researchers recently published their findings in the journal “Nature Communications,” demonstrating how onboard sensors can enable a swimming robot to learn to reach a specific destination. “Our goal is to provide physics-based guidelines that can be transferred to the design of any underwater vehicles intended for unfamiliar and diverse flow environments,” Kanso said.

Crucially, she and her lab discovered that sensing the velocity (speed and direction) of flow does not provide sufficient information for a robot to successfully navigate solo. An extra dimension of information is needed: the ability to sense flow gradients.

The flow gradient is the rate of change of the flow velocity in space. In other words, not only where or how fast the flow is going, but how quickly those factors change as the robot moves through the water.

Sensing flow gradients is the missing link in understanding how organisms are able to map and move through fluids. The paper therefore answers two questions simultaneously: not only how to enhance underwater robot-centric learning, but also how to explain why aquatic organisms are equipped with flow sensors that detect gradients.

“The clues to some of the toughest engineering puzzles are hidden in nature,” Kanso said. “Our research provides the tools to unlock that knowledge, deepening a shared understanding of the fluids that have shaped life on Earth.”