Purdue's CHORUS Center advances resilient cyber-physical systems for autonomous transportation
First NSF CISE Center led by Purdue demonstrates breakthrough technologies for safety and security in connected vehicles
As Purdue University's National Science Foundation CHORUS Center approaches its first annual site visit in February, the pioneering research center is demonstrating significant progress in making cyber-physical systems (CPS) safer and more reliable — with potentially life-saving implications for the future of transportation.
CHORUS, the first center from NSF’s Directorate for Computer and Information Science and Engineering (CISE) headquartered at Purdue, was established in 2024 with $7 million in NSF funding over five years. The center addresses a critical challenge: how to build highly dependable cyber-physical systems from components that individually may not be inherently reliable or secure.
Led by Saurabh Bagchi, professor in the Elmore Family School of Electrical and Computer Engineering (ECE) who also holds a courtesy appointment in the Department of Computer Science at Purdue, CHORUS brings together researchers from Purdue, the University of Southern California, Georgia Tech and the University of Wisconsin-Madison to tackle vulnerabilities in CPS, with an application focus on connected and autonomous transportation systems. The center's work focuses on three broad classes of challenges: naturally occurring errors in hardware and software, malevolent actors conducting security attacks, and unexpected interactions between humans and systems, which the team collectively refers to as "perturbations."
"Even when we're looking only at one vehicle, it's ripe for different kinds of failure and attack scenarios," Bagchi said. "Our main thesis is that while individual elements will fail, if one can coordinate multiple diverse assets, then it is possible to significantly reduce the incidence of failures."
One of the key challenges in autonomous transportation is that individual elements will fail, either due to natural causes, security attacks or unanticipated human interaction. In such systems, there is very little time to recover from failures, and a fallback to humans often does not solve the problem. "As a tangible example, the camera on one vehicle may be damaged, the autonomy software stack on another may be under attack, and the wireless link on a third vehicle may be degraded," Bagchi explained. "But if these three can communicate and coordinate, the individual weaknesses can be overcome. Of course, the challenge is how to determine who should coordinate, how to make the coordination fast enough for the application, and how to establish trust among the coordinating entities. We have to achieve this, all in the face of perturbations."
First-year breakthroughs
In its inaugural year, CHORUS has already achieved significant technical milestones that demonstrate the center's interdisciplinary approach and collaborative culture.
Working in coordination with the Indiana Department of Transportation (INDOT), the team has defined realistic scenarios where transportation infrastructure could be disrupted through failures or attacks, then mapped research tasks to mitigation strategies. "These application scenarios that we have jointly defined are realistic and will serve as useful guiding posts for the research of the CHORUS Center," said Nathan Sturdevant, a collaborator from INDOT.
One breakthrough technology allows the identification of objects encountered in autonomous driving by combining camera feeds from multiple vehicles. "This is advantageous because it allows the algorithm to more accurately identify objects, with inputs coming from the point of view of multiple vehicles or from cameras in the roadways," Bagchi explained. "This is also useful if perturbations affect one element in the system." This work is being presented at the 40th annual Association for the Advancement of Artificial Intelligence Conference on Artificial Intelligence as an oral in January.
Another innovation enables compute-heavy transformers — the same technology behind large language models — to run on the small devices found in vehicles. The system uses content-aware approximations, applying less aggressive simplification to complex scenes while allowing more approximation for simpler scenarios. This embedded autonomy capability was developed and tested on a physical testbed spread across several campus buildings, emulating different communication delays that vehicles would experience in real-world conditions.
The center has also developed Agile3D, a system that processes light detection and ranging (LiDAR) data directly on vehicles without needing to upload information to the cloud — a critical capability for autonomous vehicles that must make split-second decisions. This work was presented last year at the 23rd annual Association for Computer Machinery International Conference on Mobile Systems, Applications and Services.
Building infrastructure for real-world testing
Rather than rushing into expensive infrastructure investments, CHORUS took a strategic approach in Year 1. "This is so that we understand better what infrastructure elements we want to acquire and operate," Bagchi said.
The team augmented CARLA, a simulator for autonomous driving, to create scenarios with multiple vehicles and infrastructure elements — like edge compute nodes on roadways and smart traffic lights — communicating and coordinating. The center then ran a competition challenging students to find vulnerabilities that could trigger specific failure cases.
For physical testing, CHORUS established an embedded autonomy testbed with mobile graphics processing units of various generations on systems-on-chips found in vehicles of different vintages. This testbed is physically spread out over several buildings on campus, emulating different communication delays that vehicles would experience in real-world conditions.
The center has also recently gained access to a drone facility where swarms of drones can fly and coordinate among themselves. "Earlier, we had acquired and done some starting demonstrations of multiple drones flying and coordinating among themselves," Bagchi said. "With our recent access to the facility, this line of work will gather speed."
At its University of Southern California partner site, the center will acquire two Kia Nero vehicles that will be equipped with autonomy technology from another company, allowing the vehicles to achieve a certain level of autonomous operation and coordinate between themselves for testing vehicle-to-vehicle coordination capabilities
Addressing complex, real-world failures
CHORUS researchers are tackling failure scenarios that go far beyond simple technical glitches. Consider a situation extrapolated from today's infrastructure vulnerabilities: A large-scale sporting event, concert or natural disaster evacuation creates unexpected congestion in both communication and vehicular networks. The surge in traffic overwhelms available wireless bandwidth, preventing vehicles from communicating with the cloud for navigation guidance. Infrastructure cannot coordinate responsive control strategies. Then, a malicious actor launches a communication network flooding attack, further congesting the system.
Traditional methods would fall back on human handling, but if operators and drivers were relying on autonomous operation, they would be ill-prepared to take over with little notice.
"With the technology pieces we are developing in CHORUS, our modeling framework would allow us to anticipate such scenarios and then overprovision wireless network capacity, such as through a more well-resourced backhaul network," Bagchi said. "Then, during the failure situation, CHORUS' algorithms would prioritize the safety-critical traffic and make a quick determination to drop the malicious flooding traffic. Finally, it would direct the vehicles to a safe state, such as moving to a shoulder and stopping there without any accident occurring."
An interdisciplinary powerhouse
A distinctive strength of CHORUS is its interdisciplinary team, bringing together expertise from computer science, electrical engineering, economics and agricultural and biological engineering. At Purdue, the five principal investigators work closely together, often collaborating more easily than across institutions due to the physical nature of their testbeds and human user studies.
Tim Cason, Distinguished Professor of Economics in the Mitch Daniels School of Business, leads behavioral studies about how humans interact with autonomous systems while making strategic decisions. Somali Chaterji, associate professor of Agricultural and Biological Engineering, focuses on learning under noisy environments where data is incomplete or incorrect and devices have limited processing power. Shreyas Sundaram, Marie Gordon Professor of ECE, addresses control decisions that multiple entities can take in unreliable network environments to achieve goals effectively and safely. Aravind Machiry, assistant professor of ECE, develops security algorithms for small devices and uses AI to understand and patch security vulnerabilities in embedded systems. Carla Zoltowski, associate professor of engineering practice in ECE, leads curriculum development to inspire students to understand and develop solutions for cyber-physical systems (CPS) reliability and security issues.
"This interdisciplinary team and approach is a unique strength of CHORUS and this has been proven even in the very first year of its operation," Bagchi said. The team integrates distributed algorithms for low-delay coordination, control theory to guide feasible actions and avoid unsafe conditions, drone swarm algorithms adapted from agricultural monitoring, and behavioral economics principles to understand when humans are willing to concede autonomy to algorithms.
Industry partnerships drive real-world impact
CHORUS works closely with industry and government partners who are active contributors rather than partners in name only. General Motors provides expertise on autonomous vehicle characteristics at various levels of autonomy. Intel offers insights into secure embedded hardware. Amazon and Microsoft contribute knowledge about cloud and edge computing platforms where processing will run. Adobe helps develop personalized visualization methods so humans can make good decisions under time pressure. INDOT provides critical information about actual infrastructure, realistic failure situations, and policy requirements for working with vendors.
"They are not partners in name only, who write a letter at the time of the project proposal; rather, they are active and meaningful partners driving the project toward its ultimate deliverables," Bagchi said.
Applications beyond transportation
While CHORUS focuses on connected and autonomous transportation, the center's research has broad applicability to other CPS that are large-scale, involve multiple stakeholders with potentially conflicting interests, and feature interconnected elements where perturbations can cascade across organizational boundaries.
Remote-operated robotics or drone fleets, where a single operator controls multiple robots for irrigation, fire rescue or inventory management, could benefit from CHORUS innovations. So could truck platooning, where convoys of trucks autonomously follow a lead truck operated by a human driver.
Developing the next generation of CPS experts
Beyond research breakthroughs, CHORUS is committed to workforce development. The center has developed short, accessible instructional modules on topics including how big data can be used for reliability and security, reliability and security vulnerabilities of AI systems, and cybersecurity of drones.
CHORUS has also engaged students through three hackathons: Catapult, an AI and business-themed event held at Purdue in April 2025; the ML+X Machine Learning Marathon held over three months in Madison, Wisconsin, from September to December 2024; and the IEEE Computer Society North America Student Challenge Competition, held virtually from September to November 2024 with finals in Washington, D.C., in December 2024.
The center is also developing two massive open online courses — one providing foundational material on resilient cyber-physical systems and another focusing specifically on connected and autonomous transportation systems.
Looking ahead to Year Two
As CHORUS prepares for its February site visit, the team is excited to showcase live demonstrations, recorded videos and posters highlighting their technological advances. Researchers from all four partner institutions will gather with industry collaborators and NSF site visitors to demonstrate innovations like Agile3D's on-vehicle LiDAR processing and peer-to-peer learning algorithms that function even when some entities are compromised and acting maliciously.
"We are proud of the fact that we could easily give you 10 such accomplishments, most of them realized through collaboration between two or three PIs in CHORUS," Bagchi said.
In Year Two, the center will target research tasks toward specific application scenarios and realize more research elements on physical testbeds, with plans to expand infrastructure based on insights gained during the first year.
"We will move on targeting the research tasks toward the application scenarios and then realizing some elements of the research on the physical testbeds that we have and that we will be setting more of in year two," Bagchi said.
The NSF-funded CHORUS center is also a pillar of Purdue Computes, a strategic university initiative to further scale Purdue's research and educational excellence in computing and information technology.