Editor choice

2024-10-10

A new approach to protecting autonomous systems: balancing security and costs

In an era where autonomous machines are rapidly becoming a part of our daily lives, researchers have proposed a novel strategy to enhance the safety and reliability of these systems without incurring prohibitive costs. This innovative approach, developed by a team from the University of Rochester, Georgia Tech, and the Shenzhen Institute of Artificial Intelligence and Robotics for Society, promises to revolutionize how we protect robotics against vulnerabilities.

 

 

The Rise of Autonomous Machines

The landscape of autonomous technology is expanding at an unprecedented rate. Projections indicate that millions of self-driving cars will be navigating our roads by 2025, while autonomous drones are already generating billions in annual sales. This surge in adoption brings safety and reliability to the forefront of concerns for consumers, manufacturers, and regulators alike.

However, the systems designed to protect autonomous machine hardware and software from malfunctions, attacks, and failures come with their own set of challenges. These protective measures often result in increased costs related to performance features, energy consumption, weight, and the use of semiconductor chips.

 

Breaking Away from "One-Size-Fits-All"

The research team identified a critical flaw in current protection strategies: a "one-size-fits-all" approach that creates an unnecessary tradeoff between overhead costs and vulnerability protection. In response, they've proposed a more nuanced method, detailed in a paper published in Communications of the ACM.

Yuhao Zhu, an associate professor in the University of Rochester's Department of Computer Science, explained the current industry standard using Tesla as an example. "Tesla uses two Full Self-Driving (FSD) Chips in each vehicle. This redundancy provides protection if the first chip fails, but it also doubles the cost of chips for the car," Zhu noted.

The team's approach, by contrast, is more comprehensive and economical. It aims to protect against both hardware and software vulnerabilities while allocating protection resources more efficiently.

 

A Tailored Approach to Protection

"The basic idea is that you apply different protection strategies to different parts of the system," Zhu elaborated. "You can refine the approach based on the inherent characteristics of the software and hardware. We need to develop different protection strategies for the front end versus the back end of the software stack."

In autonomous vehicles, for instance, the front end of the software stack focuses on environmental sensing through devices like cameras and lidar. The back end processes this information, plans routes, and sends commands to the actuator.

Zhu pointed out that the front end doesn't require as much of the protection budget because it's inherently fault-tolerant. The back end, however, is critical to secure as it directly interfaces with the vehicle's mechanical components.

For the front end, the team suggests low-cost, software-based solutions such as filtering out data anomalies. The back end, requiring more robust protection, could benefit from techniques like checkpointing to periodically save the entire machine's state or selectively duplicating critical modules on a chip.

 

Addressing AI Challenges

The research team is now setting its sights on overcoming vulnerabilities in the latest autonomous device software stacks, which rely heavily on neural network artificial intelligence.

"Some of the most recent examples are one single, giant neural network deep learning model that takes sensing inputs, does a bunch of computation that nobody fully understands, and generates commands to the actuator," Zhu explained. "The advantage is that it greatly improves the average performance, but when it fails, you can't pinpoint the failure to a particular module. It makes the common case better but the worst case worse, which we want to mitigate."

 

Looking Ahead

This groundbreaking research, partially supported by the Semiconductor Research Corp., represents a significant step forward in the field of autonomous systems protection. By moving away from blanket solutions and towards more targeted, efficient protection strategies, the team's work could pave the way for safer, more reliable, and more cost-effective autonomous machines.

As we stand on the brink of widespread adoption of self-driving cars, autonomous drones, and other robotic systems, this new approach offers a promising path forward. It balances the critical need for safety with the practical considerations of cost, potentially accelerating the integration of autonomous technology into our daily lives while ensuring it remains as secure as possible.

Share with friends:

Write and read comments can only authorized users