Indian American researcher Rahul Mangharam leads an international collaboration to study the transition of AI agents from digital environments to the physical world, focusing on safety and cooperation.
Researchers led by Rahul Mangharam, an Indian American professor at the University of Pennsylvania’s School of Engineering and Applied Science, are exploring the implications of artificial intelligence agents transitioning from the digital realm to the physical world.
As the principal investigator of the Safe Autonomous Systems Lab (xLAB), Mangharam is spearheading a new three-year international collaboration focused on Swarm AI. This project unites three universities to examine how large teams of physical AI agents can cooperate, compete, and operate safely in real-world scenarios.
“Most of today’s AI agents live purely in software,” Mangharam explains. “We’re moving toward physical AI, systems that don’t just generate answers, but act in the real world. And once AI operates in physical space, it has to deal with real constraints and real consequences.”
Unlike their digital counterparts, physical AI agents must adhere to the laws of physics. They are required to avoid collisions, respect safety boundaries, and coordinate effectively with their teammates. The research primarily focuses on cooperation and coordination in adversarial games—situations where teams of agents must strategize against opponents while maintaining internal cohesion, as detailed in an article from Penn Today.
<p“A key technical focus is understanding intent,” says Mangharam. “Agents must infer what other agents, whether human or machine, are trying to achieve and adjust their actions accordingly. They need to coordinate without centralized control and respond to dynamic, uncertain environments. This project integrates research in machine learning for multi-agent systems that utilize game theory to facilitate cooperation and competition.”
At larger scales, the challenge becomes both algorithmic and systemic: designing distributed algorithms that can scale to tens, hundreds, or even thousands of agents, enabling them to make consistent and safe decisions in real time. A distinctive feature of this project is its emphasis on neurosymbolic AI, which merges neural networks with structured, human-encoded knowledge.
<p“You can’t just throw AI at a problem and expect it to magically figure everything out,” Mangharam asserts. “There’s always human context—hard-earned domain knowledge, engineering realities, safety rules—that doesn’t live neatly in data and can’t simply be learned from scratch. If we want these systems to work in the real world, we have to teach them the fundamentals we already understand.”
<p“By building those physical limits, safety boundaries, and operational principles directly into the system, we develop physics-informed neural networks, or PINNs, which equip AI with the necessary domain knowledge about how the world operates, the expectations, and the lines that cannot be crossed,” he adds.
In recognition of his contributions to the field, Mangharam received the 2016 U.S. Presidential Early Career Award for Scientists and Engineers (PECASE) from then-President Barack Obama for his work on life-critical systems. His accolades also include the 2016 Department of Energy’s CleanTech Prize (Regional), the 2014 IEEE Benjamin Franklin Key Award, the 2013 NSF CAREER Award, and the 2012 Intel Early Faculty Career Award. He has been selected by the National Academy of Engineering for the U.S. Frontiers of Engineering in both 2012 and 2017.
Additionally, Mangharam has received multiple best paper awards from the Association for Computing Machinery (ACM) and the Institute of Electrical and Electronics Engineers (IEEE) in areas such as Cyber-Physical Systems, controls, machine learning, and education. He earned his PhD in Electrical and Computer Engineering from Carnegie Mellon University.
This ongoing research not only aims to advance the understanding of AI agents in physical environments but also seeks to ensure that these systems can operate safely and effectively in the real world, paving the way for future innovations in autonomous technologies, according to Penn Today.