BU Team to Advance Surveillance Capabilities in Autonomous Vehicles

By Mark Dwortzan

A persistent surveillance scenario in which Smarts agents coordinate to survey a complex scene with areas weighted by importance and recognize abnormal activity.
A persistent surveillance scenario in which Smarts agents coordinate to survey a complex scene with areas weighted by importance and recognize abnormal activity.

BU test bed simulating an urban setting: Small wireless robots equipped with sensors form a team that cooperatively performs various tasks. The team automatically configures itself depending on the tasks to be performed, often reacting to random events.
BU test bed simulating an urban setting: Small wireless robots equipped with sensors form a team that cooperatively performs various tasks. The team automatically configures itself depending on the tasks to be performed, often reacting to random events.

A fleet of six unmanned Navy boats patrols a stretch of Boston Harbor for suspicious individuals, vehicles entering restricted areas and incoming liquid natural gas containers. Rigged with video cameras, laser range finders, navigation and control sensors, and on-board computers — and linked together in a network — the six vessels update one another when they detect potential security breaches. Meanwhile, a human operator continually interacts with the network to obtain critical information, but finds herself overwhelmed by the task of supervising multiple autonomous vehicles subject to ever-changing conditions.

Aiming to radically reduce the workload for human operators of semi-autonomous underwater, ground and aerial vehicles in military and civilian contexts, the U.S. Office of Naval Research launched the Smart Adaptive Reliable Teams for Persistent Surveillance (SMARTS) project on Sept. 11. Structured as a three-to-five-year Multi-University Research Initiative (MURI) and funded at $1.5 million per year, the project tasks machine learning and control theory experts from MIT, Boston University, University of California, Berkeley, and University of Pennsylvania to engineer more intelligent and capable autonomous vehicles.

Co-investigators on the BU research team are Calin Belta, assistant professor of mechanical engineering, systems engineering and bioinformatics and director of the Hybrid and Networked Systems Laboratory; and Professor Christos G. Cassandras (ECE), head of the Division of Systems Engineering and director of the Control of Discrete Event Systems Laboratory.

Using sophisticated computational tools and experimental models, Belta and Cassandras are working to develop autonomous, intelligent single agents — entities that compute, communicate and control — that can interpret and reason about their environment in changing conditions, as well as networks of multiple agents that can safely and efficiently coordinate their activities with other agents and human operators. The BU team draws on significant recent advances in robotics, sensor networks and computer, communications and control technology.

“The technology in the last 10 years has allowed us to move forward,” said Cassandras. “This includes the ability to communicate wirelessly across many agents or robots, and to pack much more powerful computational capabilities in smaller spaces.”

He compared the project’s main challenges to those posed by the “traveling salesmen” problem, in which a group of salesmen must visit several cities in minimum time.

“You have a large number of cities, some more important than others, some appear and disappear, the salesmen may lose communication with each other, their cars may break down and they have to visit as many cities as they can within a set time,” Cassandras explained. “How do you coordinate all of this? At the highest level, you want to define the task in simple English so the team can efficiently decipher the details.” 
   
In the SMARTS project, each vehicle or agent must make decisions with minimal input. Ideally, technology conceived by project researchers will enable semi-autonomous vehicles to make decisions completely independent of human interaction except when absolutely necessary — regardless of changes in weather, lighting or other ambient conditions.

In the military theater, the ultimate goal of this research is to create teams of persistent surveillance agents to give combat vehicles the edge in detecting and responding to hostile targets.

“Their missions are to go, detect, visit targets and come home,” said Cassandras.

SMARTS technology may also be developed to enable motor vehicles to carve out an optimum path to an empty parking spot through congested traffic; empower sensor networks to display and control energy consumption by dishwashers, washer-dryers and other home appliances; and provide health monitoring services to homebound elderly or incapacitated adults.
 
With these applications in mind, Belta and Cassandras are working both on computers — using optimization methods to model the behavior of single and multiple agents and probabilistic techniques to model uncertain conditions — and deploying small robots that carry cameras, communicate with each other and perform missions in simulated settings.

For instance, to test out autonomous agent decision-making performance in an urban context, Belta sends robotic cars on various missions in a model city graced with plastic towers, makeshift roads with parking spots and computer-controlled traffic lights.

“We’re trying to come up with formal proofs for our controls and communications strategies and to ensure they’re bug-free,” he said. “We want to make sure that our control systems always work, regardless of operating conditions.”
 

Back to news headlines