Creating the Next Generation of Connected Autonomous Systems: an Interview with Weisong Shi on Edge Computing, Autonomous Driving, and the Future of Mobility

An interview with Weisong Shi, recipient of the 2026 IEEE Computer Society Edward J McClusky Technical Achievement Award.

Weisong Shi is the Alumni Distinguished Professor and Chair of the Department of Computer and Information Sciences at the University of Delaware, whose pioneering vision for edge computing and autonomous mobility has redefined the landscape of connected infrastructure.

and security.

We connected with Dr. Shi to discuss the transition from cloud to edge, the challenges of securing autonomous vehicle networks, and the integration of smart health technologies into everyday life.

Your paper "Edge Computing: Vision and Challenges" has over 10,000 citations and essentially helped define a subfield. For a student today who sees "Edge" as a buzzword, how would you describe the fundamental shift in thinking required to move computation from a centralized cloud to the periphery?

This is a great question. It was a bit of a challenge to sell the concept of “Edge” 10 years ago, but today it is much easier, as we have seen many urgent needs. The fundamental shift isn't just about where the computer sits, but who the data serves and how fast it needs to act.

For decades, we were trained in a 'Cloud-Centric' mindset: collect data at the source, ship it to a giant data center, process it, and send a result back. That works for some type of applications, e.g., movie recommendation, but it fails for an autonomous car or a surgical robot.

The shift to the 'periphery' requires three major changes in thinking:

• From Bandwidth to Latency: In the cloud, we care about how much data we can collect. At the edge, we care about how quickly we can react. We had to start thinking about 'deterministic' response times, especially in the physical world.
• From Abundance to Scarcity: The cloud feels infinite. The edge is defined by constraints—limited power, limited cooling, and limited compute. Thinking 'Edge' means learning to be computationally elegant again.
• From Privacy by Policy to Privacy by Design: Moving computation to the periphery means the data never has to leave its owner. It changes privacy from a legal promise made by a cloud provider into a physical reality of the architecture.

To a student today, I would say: Edge computing is about moving the 'brain' as close to the 'eyes' and 'hands' of the system as possible.

You focus on connected and autonomous vehicles. What is the most significant "bottleneck" currently preventing the large-scale AV deployment—is it a lack of raw processing power, or the latency in how vehicles communicate with each other?

Actually, the most overlooked bottleneck isn't just a number like '5 milliseconds'—it is the tight coupling between the processing latency and the physical velocity of the vehicle.

People often forget that in autonomous driving, time is literally distance. If a vehicle is traveling at 65 mph (approximately 29 meters per second), a 100ms delay in processing or communication isn't just a 'laggy' experience—it means the vehicle has traveled nearly 3 meters (10 feet) before it even begins to react. The real bottleneck, therefore, is achieving Deterministic Latencies at different speeds. We need a guarantee that the system will respond within a specific timeframe, every single time, regardless of network congestion.

In addition to computation and communication latency, factors such as weather conditions and vehicle status significantly impact performance. To address these variables, our recent work introduces DaVoS: a physics-based, platform-independent safety framework. By calculating a real-time safety-confidence score from five key metrics—including predicted conflict distance, motion state, variability, system overhead, and speed—the DaVoS model facilitates the design of adaptive autonomous driving systems ready for large-scale deployment.

You served as an NSF Program Director and now chair a major department at the University of Delaware. What did your time at the NSF teach you about how "big ideas" get funded that you didn't realize when you were just a junior faculty member?

My time at the NSF taught me a fundamental truth that many junior researchers miss: Research papers are for selling solutions, but proposals are for selling problems.

When you write a paper, the community expects you to show exactly how you solved a specific challenge with proven results. However, when you write a proposal—especially for a 'Big Idea'—you are asking the community to agree that a specific problem is worth solving and that it is the right time to tackle it.

Here is how that shift in mindset changed my approach:

• Defining the New Frontier: If the problem you are attacking is genuinely new, or if you’ve identified a 'blind spot' in the current landscape, the proposal will almost naturally achieve community resonance.
• Resonance through Urgency: A great proposal convinces the reviewers that if we don't solve this specific problem now, the entire field will stall. You are essentially 'selling' the gap in our knowledge.
• The Solution is a Promise, the Problem is the Reality: While you must provide a credible path to a solution, the 'Big Idea' resides in the formulation of the question. Once the community buys into the importance of your problem, they become invested in seeing you succeed.

Learning to 'sell the problem' is what allows a researcher to move from incremental work to field-defining leadership.

How do you balance the immediate, bottom-line needs of industry partners with the long-term, "blue sky" goals of academic research?

As a computer scientist, I believe our most important job isn't to compete with industry on what they are building today, but to anticipate what they will desperately need five years from now. If you are working on the same problem as a hundred-person engineering team at an OEM or AV startup, you’ve already lost. You have to look for the 'silent' bottlenecks that only appear at scale.

A perfect example from our recent work is Data Storage Systems for Autonomous Vehicles (AVS). Right now, most OEMs and startups are focused on the 'brain'—how the car perceives and decides. Very few are looking at the 'memory'—how that vehicle efficiently stores, logs, and retrieves the terabytes of data it generates every single day.

• The Impending Crisis: Today, a single test vehicle can generate over 15-32 TB of data per day. When we move from a few dozen prototypes to a fleet of 10,000 deployment-ready vehicles, the current 'ad-hoc' storage methods will collapse.
• The 'Day 2' Problem: Industry will soon face massive challenges in real-time append logging, queryable historical data for accident analysis, and hierarchical storage that decides what to keep and what to discard.
• The Research Advantage: We’ve developed AVS (Autonomous Vehicle Storage) and integrated it into our DAVOS operating system specifically to address this. We aren't just building a hard drive; we are building a hierarchical, queryable management system that treats a vehicle like a mobile data center.
• By the time the industry realizes their storage architecture is the bottleneck, we will already have the foundational solutions ready for them. That is the true value of academic research: solving the problems that industry doesn't even know it has yet.

As the Editor-in-Chief of Smart Health and a founder of the CHASE conference, you bridge two very different worlds. What project are you currently working on in this space?

One of the most rewarding projects we are currently leading at the CAR Lab is the development of an autonomous wheelchair. For us, this isn't just a 'robotics' project; it is the ultimate expression of Physical Intelligence on the Edge.

What makes a wheelchair a unique challenge compared to a car is the environment. A car stays on a road with lanes and signs. A wheelchair must navigate crowded hospital hallways, interact with elevators, and respond to spoken commands—all while ensuring the safety of a vulnerable user.

• In-Situ Intelligence: Just like our work with AVS for cars, the wheelchair must process its surroundings locally and instantly. We’ve developed a prototype that combines onboard sensors with software that interprets natural language, allowing a user to simply say 'take me to the cafeteria.'
• Revolutionary Healthcare Integration: This ties directly into our vision for Smart and Connected Health. By turning the wheelchair into a mobile edge node, it can continuously monitor the user's vitals and provide real-time feedback. If the system detects a health emergency, the 'vehicle' doesn't just stop—it can automatically navigate to the nearest medical professional or alert the hospital's emergency system.
• Mobility for All: This project embodies our long-term goal of 'democratizing' autonomy. We are moving toward a future where mobility is no longer a privilege of the able-bodied, but a guaranteed right supported by invisible, intelligent edge infrastructure.