Yen-Ling Kuo Advances Robotic Autonomy Through Uncertainty Estimation

Yen-Ling Kuo, an assistant professor of computer science at the University of Virginia, is redefining how autonomous systems manage uncertainty. Her research, specifically the development of the Diff-DAgger framework, enables robots to identify when they are operating outside their training data, allowing them to solicit human intervention only when necessary. This approach has demonstrated a 39 percent improvement in failure prediction and a 20 percent increase in task completion rates, according to data published by the IEEE Robotics and Automation Society.

How Diff-DAgger Reduces Human Supervision

Traditional robot learning often relies on human operators to manually guide machines or provide constant monitoring, which becomes inefficient as environments change. The Diff-DAgger method addresses this by repurposing “diffusion loss”—the signal robots use to improve models during training—as a real-time confidence metric. When a robot encounters an unfamiliar scenario, this signal spikes, indicating that the machine cannot reliably predict the outcome. By automating this “self-diagnosis,” the system reduces the need for constant human oversight, allowing robots to function independently during routine tasks and only triggering alerts during high-uncertainty events.

The Intersection of Cognitive Science and Robotics

Kuo’s approach is rooted in her interdisciplinary background, which combines computer science with cognitive science. Before joining the University of Virginia in 2023, Kuo completed her Ph.D. at MIT, where she worked within the Computer Science and Artificial Intelligence Laboratory (CSAIL). Her research frequently incorporates the “theory of mind” concept, which allows robots to infer intent and mental states from human signals like gaze, movement, and language. This framework, supported by a National Science Foundation CAREER Award, aims to move beyond simple task mimicry toward systems that can interpret social and physical cues in complex, shared environments.

From Industry Experience to Academic Innovation

Kuo’s technical perspective was shaped by four years as a software engineer at Google, where she worked on computer vision and natural language processing. During her tenure, she led the “Shop the Look” initiative, which integrated neural networks into shopping search experiences. This professional experience highlighted a critical gap in deep learning: the lack of transparency in how neural networks arrive at specific decisions. This realization drove her return to academia to study the fundamental mechanics of machine intelligence, eventually leading to her current work in the UVA Engineering Link Lab.

Future Directions in Human-Robot Interaction

As autonomous systems become more integrated into daily life, Kuo’s work focuses on making these interactions more intuitive. Beyond her work in robotic manipulation, she is exploring how self-driving vehicles can better reason about road interactions and driver behavior, an effort recognized by the Toyota Research Institute’s Young Faculty Researcher Award. Her ongoing research aims to build robots that do not just perform programmed sequences but actively interpret the social context of their surroundings, bridging the gap between machine logic and human-centric environments.

Key Research Milestones

• 2022: Completed Ph.D. in computer science at MIT, focusing on AI systems that apply past learning to new situations.
• 2023: Joined the University of Virginia as an assistant professor; received the IEEE-RAS Outstanding Women in Robotics and Automation Early Career Contribution Award.
• 2024: Awarded a $665,000 NSF CAREER grant to develop computational models for theory of mind in human-robot interactions.