Endless learning

bol30 mat02

With the aim of increasing the safety of drivers around the world, Volvo assembler has adopted state-of-the-art technology and started to apply sensors in its vehicles to evaluate the most different situations, from detecting possible malfunctions of automobiles to creating warnings about risks of collisions. To obtain accurate data, such equipment records a multitude of information inside and outside the car, such as oil pressure or even detection of objects around.

The use of sensors is not exactly new to the automotive industry. What Volvo wanted to do was take a step forward and implement a system that can learn according to use, in a process very similar to what happens in human learning. To that end, it formalized a partnership with the Federal University of São Carlos (UFSCar), which is proud to have created the first endless learning system in computing history, NELL, in partnership with the North American Carnegie Mellon. Now the system is being taken to the automotive segment.

"The analysis of sensor data is similar to what happens on an online platform like Netflix. After you watch a movie or a series, an algorithm detects a pattern and starts to suggest attractions that are according to your preferences”, explains Professor Estevam Hruschka, who heads the project within UFSCar.

The point is that the algorithm of Netflix and several other similar platforms is static, while what is being developed by the partnership between the assembler and the educational institution is constantly evolving. Exemplifying: If a data set indicates motor overheating and the driver verifies that it is only a false alarm, it will record this information to the system. Automated learning will allow the car's computer to learn from it, not to issue the alert again in a similar situation.

"With practice, the system will also become 'accustomed' to user preferences, understanding the situations in which he would like to receive audible or visual alerts. In some cases, such signals can be harmful, as in times that require a lot of concentration, and the system will adapt to that", he explains. The system is complex and will also prevent mechanical failures, material fatigue and various other situations, all from sensor data measurement.

The cooperation work took place between February and December 2017 and involved the visit of a Brazilian researcher to Sweden, Amandia Oliveira Sá, further tightening the ties between the institutions. "At UFSCar, we have found a competent partner with an international approach”, says Anders Hedebjörn, Volvo's Senior Manager R & D Projects in Brazil. For him, "the horizon of Brazilian researchers is highly relevant, which allowed the creation of knowledge about future solutions. This has given us an understanding and direction on what we can develop and implement to maintain our leading position in the segment over time".

The gains were mutual. Professor Hruschka points out that the proximity to the industry was a huge prize for researchers at UFSCar. "In a research lab, it is possible to obtain a series of data. However, having a partner like Volvo in which you can get real-life data on equipment usage, as well as exchanges of experiences with the engineers who developed the sensors, information on what happens in the car and other data, is a fabulous breakthrough for scientific research", he says.

As much as the partnership has come to an end, the outlook is one of continuity. There are still conversations between institutions where long-term planning is being defined for even more gains for both parties.