An interdisciplinary research team from the University of Waterloo is building systems that mimic the human brain to improve the power efficiency and performance of artificial neural networks like those used in autonomous vehicle (AV) technology.
Self-driving cars must navigate, make decisions and react in real time to changing environments, and while artificial intelligence (AI) has made significant strides in this area, road safety remains a concern. To improve how an AV performs complex tasks, such as its response to a critical situation, requires supplying its AI system with a lot of computing power. This increased drain on the vehicle's battery could impact its driving range.
Dr. Chris Eliasmith, a professor jointly appointed to the Departments of Systems Design Engineering and Philosophy, leads Waterloo's Computational Neuroscience Research Group (CNRG) that focuses on replicating human brain function to create more efficient and powerful artificial systems.
"The brain is by far the best autonomous system we know," Eliasmith says. "It operates with unmatched power efficiency, using only about 20 watts. Computers are smart and AI language models like ChatGPT are very human-like, but they use at least 1000 times more power, which makes them impractical for extensive mobile use.
"If we can take what the brain does naturally and apply it to AVs, we can build autonomous cars that not only think faster and better but also conserve battery power, allowing for better overall performance," Eliasmith adds.
Building smarter, safer cars
In collaboration with Mercedes Benz, CNRG will apply their neuromorphic computing expertise - designing and developing software and hardware development designed to mimic how the brain works - to making autonomous vehicle technology safer and more efficient. This collaboration highlights the University's commitment to building meaningful industry and research partnerships for societal, economic, technological, health and sustainable impact.
AV systems struggle with complex tasks like "scene understanding," which Eliasmith explains is the use of body language and eye contact to interpret whether a pedestrian is about to cross the road. Using simulations and neuromorphic technology, the lab will enhance the system's perception, prediction and control features, improving its ability to read and react to its environment correctly.
Dr. Chris Eliasmith, University of Waterloo professor jointly appointed to the Departments of Systems Design Engineering and Philosophy, and head of the Computational Neuroscience Research Group (CNRG).
"Our research using neuromorphic computers has already demonstrated a 10 to 100 times reduction in the amount of power required to do control and perception tasks without any loss of accuracy, and often with improvements," Eliasmith says.
"We believe these and other benefits, like enhanced responsiveness, will scale well to fully autonomous driving systems, making them more robust and ultimately safer, while using far less power."
Innovation on and off the road
Eliasmith's work in neuromorphic computing exemplifies Waterloo's interdisciplinary and forward-thinking approach to solving complex challenges. His research extends beyond AVs into areas like household devices and wearables, all designed to perform complex tasks with minimal energy consumption. His spin-off company, Applied Brain Research (ABR), is developing the world's first AI chip that can support full vocabulary speech recognition at extremely low power.
"Imagine a kitchen appliance that responds to unscripted, natural voice commands to defrost a chicken or preheat the oven," Eliasmith says. "Or conversing with any disconnected device, like a virtual reality headset, in the same way you would with a person - all using a fraction of the power required by today's AI systems.
"The CNRG lab and ABR are both working on exciting applications of neuromorphic computing that will transform how we live and move about in the world. The human brain is awesome technology - why try and reinvent the wheel?"
