Scientists have crafted a dynamic computer program aimed at increasing traffic safety by anticipating moments when drivers are in an optimal state to engage with vehicle-integrated systems and accept notifications, such as traffic updates, incoming communications, or navigational instructions.
The team from Cambridge University, in collaboration with Jaguar Land Rover (JLR), utilized a blend of real-world driving tests and advanced computational models, including Bayesian filtering methods, to consistently and precisely gauge the level of attentional demand, or “workload,” on the driver. Navigating unknown territories might represent high demand, whereas routine travel might indicate less demand.
The innovative algorithm adapts and reacts swiftly to shifts in driving behavior and environment, including variability in road conditions, road design, or individual driver traits. This gathered data can be seamlessly integrated into a car’s internal systems like entertainment interfaces, GPS, displays, and advanced driver assistance systems (ADAS), among others.
Vehicle-driver interactions can consequently be tailored to prioritize safety and improve overall enjoyment, providing adaptive interactions between the driver and the machine. For instance, notifications are scheduled for moments when the driver is under less strain, allowing them to maintain concentration on the road during tense driving intervals. These findings are documented in the “IEEE Transactions on Intelligent Vehicles” journal.
“As drivers are increasingly bombarded with data, this could become a significant hazard for road safety,” stated Dr. Bashar Ahmad, co-lead author from the Department of Engineering at Cambridge. “While vehicles have the capacity to share a multitude of information with the driver, it’s not always secure or feasible unless the driver’s condition is known.”
A driver’s mental load can vary often; challenging driving scenarios, such as navigating through high-traffic areas or adverse road conditions, are typically more strenuous than routine commutes.
“Receiving a message in a taxing driving situation could be quite inconvenient,” said Ahmad. “The obstacle faced by automobile producers lies in identifying the moments when a driver is sufficiently at ease to interact with the vehicle and receive notifications without causing a distraction.”
Existing algorithms that assess driver demand utilize tools like eye-tracking devices and heart rate monitors to gather biometric information. The Cambridge team was interested in creating a method that could achieve similar results using basic vehicle data, like steering, acceleration, and braking inputs. This method should also be capable of integrating and synchronizing various data streams with differing update frequencies, and even biometric readings if these are available.
To assess the driver’s mental load, the scientists initially designed an automated version of the Peripheral Detection Task to collect subjective workload information during the drive. A smartphone displaying a navigational map was attached to the car’s central vent alongside a small LED light that blinked periodically. All participants drove the same route encompassing rural, urban, and major roads. They were instructed to press a button worn on the finger each time the LED flashed red, and they felt the mental demand was low. Video evaluation of these driving sessions, combined with button response data, enabled researchers to pinpoint high-demand situations, such as congested crossroads or erratic behavior from surrounding vehicles.
Following the road tests, the collected data was employed to establish and fine-tune a supervised machine learning model capable of profiling drivers by their average workload experience, and a flexible Bayesian filtering method for instantly estimating the driver’s real-time workload using multiple vehicle performance indicators, such as steering and braking patterns. The approach creates a comprehensive view of workload by considering both long-term average profiles and instantaneous assessments.
“For most machine learning setups like this, specific driver training would be required, but we’ve managed to adapt the models spontaneously using straightforward Bayesian filtering methods,” explained Ahmad. “It efficiently adjusts to various road styles and conditions, as well as different drivers in the same vehicle.”
This venture was performed in partnership with JLR who led the experimental design and data acquisition. It is part of a JLR-sponsored project under the agreement with Cambridge University.
“This research is critical in comprehending the influence of our design from the user’s perspective, thus enabling us to constantly refine safety and provide our customers with unique driving experiences,” remarked Dr. Lee Skrypchuk, JLR’s Senior Technical Specialist for Human Machine Interface. “These insights are set to influence how we implement intelligent scheduling in our vehicles so that drivers receive pertinent notifications at optimal times, promoting smooth and stress-free travel.”