Professor and Director of the Stevens Institute for Artificial Intelligence (SIAI) Brendan Englot has received a $1.9 million grant from the U.S. Navy. With this support from the Office of Naval Research, Englot will spearhead a pioneering five-year project to enhance the autonomy and operational capacity of uncrewed surface vehicles (USVs). His work targets improving these autonomous vessels’ capabilities in tackling complex, often unpredictable missions in marine environments.
The study, “Advanced Autonomy for Unmanned Surface Vehicles (USVs) via Distributional Reinforcement Learning,” seeks to take USV autonomy to new levels, particularly in scenarios requiring rapid adaptation and decision-making. Englot’s team will investigate the potential of distributional reinforcement learning, a sophisticated method that builds upon traditional reinforcement learning. This approach equips USVs with more comprehensive situational awareness and enables them to make informed, safety-focused decisions in real-time.
“Think of it like autonomous whitewater rafting,” Englot explained. “Even with maps and GPS, a vessel may not be aware of threats such as boulders, eddies, or shifting currents until they’re encountered. We aim to develop methods for USVs to learn safe navigation in challenging, dynamic conditions to complete their missions effectively.”
Englot’s vision is to expand USVs’ applications beyond their current uses, focusing on relatively straightforward monitoring tasks. His team envisions USVs as central players in crucial missions such as tracking algal blooms and oil plumes, assessing storm damage, assisting in search-and-rescue efforts, and working with aerial drones and other marine vehicles for complex, coordinated operations.
Englot’s prior Navy-funded research on enhancing autonomous vehicle reliability inspired this new project. Drawing on his mechanical engineering and artificial intelligence background, he aims to push the boundaries of what crewless marine vessels can accomplish. His methodology revolves around distributional reinforcement learning, which allows USVs to account for potential outcomes with varying degrees of uncertainty, unlike traditional reinforcement learning, which only considers expected outcomes.
The project will unfold in three primary stages. During the initial stage, Stevens Ph.D. students will develop systems that allow USVs to tackle diverse and demanding tasks in unpredictable conditions. In the second stage, the team will focus on managing multiple USVs as a coordinated fleet, enabling them to work together effectively. Finally, the vehicles will undergo rigorous testing, moving from basic 2D and 3D simulations to real-life environments.
One unique aspect of this research is Englot’s approach to training these vehicles in low-fidelity and high-fidelity simulations. These dual training environments maximize the adaptability and effectiveness of USVs in real-world missions, paving the way for safer, more efficient fleet operations.
Englot’s project will also foster collaboration with other Navy-funded researchers and naval personnel who bring valuable insights into mission needs and strategic challenges. This collaboration enhances the research’s applicability and could accelerate the deployment of USVs in critical Navy missions.
“I hope that our research will allow USVs to perform key tasks in mission-critical situations in the most challenging environments,” Englot said. “Beyond advancing this essential technology, it also promises to improve our understanding of coastal ecosystems and infrastructure and, ultimately, keep more humans out of harm’s way.”
As the director of the Stevens Institute for Artificial Intelligence, Englot’s work embodies Stevens’ commitment to innovation and real-world problem-solving. This ambitious project could revolutionize the Navy’s ability to deploy USVs, making coastal and marine operations safer and more efficient. Through projects like these, Stevens continues to play a vital role in advancing autonomous technology for critical national and environmental needs.
Courtesy of stevens.edu
