Anca Dragan: Engineering Trust in Human-AI Collaboration — Innovations in Robotic Safety and Value-Aligned Autonomy

Early Life and Education

Anca Dragan was born in 1986 in Bucharest, Romania, during the final years of the country’s communist regime. Growing up in a society transitioning to democracy, she developed a fascination with problem-solving through mathematics, a discipline she viewed as a universal "language of logic." Her precocious talent emerged early: by age 14, she ranked among Romania’s top performers in national physics and mathematics competitions, eventually representing her country at the International Mathematical Olympiad. Dragan credits her rigorous Romanian education system—known for its emphasis on abstract reasoning—with sharpening her analytical mindset.

She pursued undergraduate studies at the University of Bucharest, graduating summa cum laude with a B.Sc. in Mathematics in 2005. Dragan’s interest in applied systems led her to Carnegie Mellon University (CMU), where she earned an M.S. in Robotics in 2007 under the mentorship of roboticist Siddhast Srinivasa. Her thesis explored early concepts of human-robot motion planning, foreshadowing her later focus on collaborative autonomy. Dragan remained at CMU for her Ph.D., completing her groundbreaking dissertation, Motion Planning for Robots in Human Environments, in 2015. The work formalized algorithms for robots to predict and adapt to human behavior—a foundational step toward safe human-AI interaction.

Career and Research Contributions

Appointed as an assistant professor at UC Berkeley in 2017, Dragan established the InterACT Lab (Interaction and Collaboration with Autonomous Systems), a hub for redefining how robots and AI systems coexist with humans. Her research pivots on two pillars: value alignment (ensuring AI objectives match human intent) and legible autonomy (designing systems whose actions humans can interpret). A seminal contribution is cooperative inverse reinforcement learning (C-IRL), developed with Pieter Abbeel and Stuart Russell. Unlike traditional IRL, which assumes passive human observation, C-IRL frames human-AI interaction as a dynamic game, enabling robots to infer intentions while transparently signaling their own goals. This framework now underpins collaborative robots in healthcare (e.g., assistive exoskeletons) and manufacturing.

Dragan’s work extends beyond academia. As a consultant for Waymo (Google’s self-driving division), she co-designed safety protocols for autonomous vehicles navigating unpredictable pedestrian behavior. Her 2020 paper, "The Hidden Cost of Simplicity: Why Reward Engineering Isn’t Enough for Aligned AI," challenged overreliance on predefined reward functions, advocating instead for models that learn ethical constraints through human feedback—a concept central to Berkeley’s Center for Human-Compatible AI (CHAI), where she serves as a lead researcher.

Her interdisciplinary approach blends control theory, cognitive science, and ethics. Notable projects include:

• Shared Autonomy Algorithms: Enabling drones and surgical robots to adjust autonomy levels based on human expertise.
• Prosocial Robot Behavior: Developing robots that prioritize societal norms (e.g., queuing protocols) in public spaces.
• AI Transparency Tools: Creating visualization systems to explain autonomous decisions in real time.

Dragan frequently emphasizes that "safety is not a constraint but a core design principle," a philosophy reflected in her advocacy for regulatory frameworks governing AI deployment.

Awards and Honors

• Sloan Research Award 2017
• 2020 MIT TR35 (Innovators Under 35)
• 2021 IEEE Robotics and Automation Early Career Award
• 2022 AAAI Early Career Award
• NSF CAREER Award (2019)