Clark to enhance safety of autonomous systems, even when under attack

Andrew Clark received a National Science Foundation grant to ensure the safety and reliability of drones, driverless cars and surgical robots

Shawn Ballard 01.14.2025

Autonomous vehicles are susceptible to sensor malfunctions and mechanical failures that might make them unable to detect and appropriately respond to real-time conditions. They also face the threat of cyberattacks in which bad actors might spoof the car’s location or fabricate the presence of an obstacle, causing the vehicle to behave dangerously. (Image: iStock)

Facebook Twitter Linkedin Email

Ensuring the safety of autonomous systems, such as driverless cars, unmanned aerial vehicles and surgical robots, is a critical challenge in the growing field of automation. A new award supports work at the McKelvey School of Engineering at Washington University in St. Louis to develop a framework that will allow these systems to maintain safety even in the face of sensor malfunctions, mechanical failures or deliberate cyberattacks.

Andrew Clark, associate professor in the Preston M. Green Department of Electrical & Systems Engineering, received a $454,202 grant from the National Science Foundation (NSF) to support his research on safe control of autonomous systems in adverse conditions. Whether malfunctions result from naturally occurring faults or deliberate attacks, Clark aims to combine techniques from control theory, machine learning and system security to guarantee system functionality across a wide range of autonomous systems and fault or attack scenarios.

“My focus in this project is on creating safety filters that can constrain the behavior of autonomous systems in real-time, ensuring safe operation even when key components of the system are compromised,” Clark said. “These safety filters will be part of a unifying decision-making framework designed to detect critical failures and respond with tailored safety measures, while also maintaining flexibility.”

One of the primary challenges Clark will address in this project is the vulnerability of learning-enabled systems – those powered by machine-learning algorithms – which often struggle to adapt when confronted with novel situations not seen during training. For example, if a self-driving car encounters a GPS disruption it was not trained to recognize, its performance could degrade significantly, leading to unsafe behavior. Ultimately, Clark’s research seeks to enhance the reliability of autonomous systems in unpredictable environments, supporting widespread deployment of autonomous and learning-enabled systems across a variety of real-world applications.