I. The story of Safety Feed
Around five years ago when we started designing the world’s first in-car voice assistant, our primary objective was to reduce the number of accidents caused by drivers being distracted by their mobile devices. We believed that voice assistance would allow for safe and easy device interactions, keeping drivers’ hands on the wheel and eyes on the road.
However, during our collaborative study with the Institute for Applied Psychology, mensch-verkehr-umwelt (mvu), we confirmed that a constant flow of information is distracting and can lead to driver errors— even for experienced drivers.
Our voice assistant faced a challenge that any good human co-driver can sense instinctively: knowing when to keep quiet. If the assistant voices an incoming message while a driver attempts to navigate a complex roundabout, it puts cognitive strain on the driver and reduces driving performance, leading to an increased error rate.
II. Cognitive Load, and why it matters
At any given moment on the road, a professional driver is subject to a number of factors that affect their driving performance. We measure these factors in terms of ‘cognitive load’, which we further categorise according to complexity and risk. Adjusting navigation, responding to dispatch, racing against a tight delivery schedule, or in the worst case, sending a text message — these situations all add varying degrees of pressure to an individual’s cognitive load. Some of these situations are also illegal, but unfortunately not unusual.
Our objective was to build a more intelligent voice assistant that would be able to assess the cognitive load on the driver and to deliver voice notifications only during opportune moments, thus freeing the driver up to perform their tasks safely while reducing their error rate.
III. How it works: The SAFE index
Safety Feed is built around our voice adaptation of mvu’s proprietary SAFE index, a classification system for upcoming driver hazards. We had to try many different approaches to create these classifications.
In the example below we see a standard section of urban road in Berlin with a 50 km/h speed limit. This is a relatively low complexity situation for the driver. It is not so tedious as to create an underload (imagine a long, empty stretch of highway at night), but it has just the right amount of complexity where we can deliver a voice notification without endangering the driver. This easy section is demarcated in green and scores 82 for complexity on our SAFE index.
However when the driver approaches a complex situation, the system knows not to deliver voice notifications in order to keep the cognitive load as low as possible and avoid the potential for overload. Essentially, the last thing a driver needs while attempting to navigate an unfamiliar situation is an assistant reading out a lengthy voice prompt.
This is represented below by the red line crossing the busy urban roundabout, giving a SAFE index score of 110 for complexity, thereby delaying the voice notification until a safer moment arrives.
In order to establish the criteria necessary to judge cognitive load, we use a combination of artificial intelligence and rules-based filtering. This involves enriching a complexity assessment from 126 defined traffic situations, with real-time data such as weather and time of day. The result has a direct influence on the dialogues of the speech assistant, which, just like a real passenger, understands when it is better to remain silent.
IV. Putting it to the test
Having received funding from the Federal Ministry of Transport and Digital Infrastructure (BMVI) as part of the mFUND program, we could build and begin to test Safety Feed.
Testing began in Munich, in June and July of 2020. Over the course of 60 test drives lasting several hours in rush hour traffic, messages were played back via smartphone in high-complexity situations. Our cohort of test drivers, which contained many experienced drivers, made a total of 547 errors without Safety Feed.
This result improved significantly with the introduction of Safety Feed. The messages from the smartphone were suppressed in complex situations and played at a more appropriate moment. The result: Fewer speed violations, less unadjusted acceleration, fewer priority and subordination errors and less obstruction or danger to pedestrians and cyclists. “The effect is highly significant,” confirms Prof. Wolfgang Fastenmeier, head of the mvu Institute.
Just over half of the test subjects reported that they feel distracted by messages on their smartphones when manoeuvring through “tricky” situations, such as turning at intersections or driving through narrow streets. Just under a third said that these messages affected their performance while driving. Almost three quarters said that they would prefer not to receive the messages at intersections.
V. Tomorrow’s fleets: Safer and smarter
Next up is to integrate Safety Feed into German Autolabs’ voice assistance platform for professional drivers. Our platform augments drivers from logistics and transport with voice assistance powers to make their working lives safer, smarter and easier.
“Professionals behind the wheel in last mile delivery, long-distance travel or sales are particularly challenged in dense traffic,” concludes Holger G. Weiss, founder of German Autolabs. “Intelligent voice control makes these fleets safer and more productive.”
German Autolabs builds voice assistance for professional drivers. Our platform is customizable, easy to deploy and brings voice powers to fleets around the world.
To find out more, visit germanautolabs.com — and thanks for reading.