Harvard SEAS Research Targets 'Uncertainty' to Move Autonomous Systems Into the Wild

Assistant Professor Stephanie Gil at Harvard's John A. Paulson School of Engineering and Applied Sciences (SEAS) is developing new methods to enable autonomous systems to operate effectively in complex, real-world environments. Her research focuses on improving decision-making and coordination for robots, drones, and vehicles operating in scenarios where information is incomplete, unknown, or untrustworthy.

According to Harvard SEAS, the primary barrier to deploying these systems outside of controlled settings is "uncertainty." This includes challenges such as navigating entirely unknown environments, dealing with incomplete sensory data, and managing untrustworthy network data. To solve this, Gil is combining classical decision-making frameworks with AI-driven pattern recognition, aiming to transition autonomous agents from laboratory settings into "the wild."

The Shift Toward Networked Coordination

While autonomous vehicles have already seen deployment for ridesharing in various cities, they currently operate largely as individual agents. Gil's work seeks to evolve this model into a networked system of physical agents—including boats, drones, and robots—that can coordinate actions in real-time across urban or remote landscapes.

As part of this evolution, Gil notes that the next phase for autonomous rideshare vehicles involves deeper coordination between the vehicles themselves and the surrounding infrastructure. For example, integrating vehicle systems with intersection cameras could allow the network to anticipate demand and reduce wait times more effectively. "The next phase is mobile systems that are also connected to each other, but that can act physically in the world," Gil stated.

High-Stakes Implications

Improving decision-making under uncertainty is critical for the safe deployment of autonomous systems in high-stakes environments. The research has direct applications for search and rescue operations, deep-sea exploration, outer space missions, and urban mobility, where a failure in judgment can have severe physical consequences.

Beyond simple navigation, a key focus of the research is the security and resilience of these systems. Ensuring data integrity is essential to mitigate the risks associated with hacked or poisoned data. Gil is working to create algorithms that "break gracefully," ensuring that when a system encounters corrupted data or an impossible scenario, it does so in a way that minimizes risk and maintains safety.

Future Outlook

As these systems move toward broader integration, the focus will remain on the intersection of connectivity and physical action. The ability for diverse agents to share untrustworthy or incomplete data and still reach a safe, coordinated decision remains the central challenge. Future developments will likely center on how these networked agents maintain stability when the external environment becomes unpredictable.