Seven seconds before it happened. BMW develops artificial intelligence that predicts accidents
BMW researchers have developed a new intelligent vehicle early warning system capable of learning from thousands of real traffic scenarios.
According to a new study carried out with BMW Group, published in the journal IEEE Smart Transport Systems, you can soon ride a self-driving car and press the ai driving button to survive fatal accidents because you will be warned 7 seconds before the accident, which is enough time to make quick decisions that can save thousands of lives.
According to the study, whose results were reviewed by interesting engineering, Artificial Intelligence can make these predictions with more than 85% accuracy, with the AI early warning system predicting critical scenarios.
The research team teamed up with Professor Ekehard Steinbach, Head of Media Technology and Board Member of the Munich School of Robotics and Machine Intelligence (MSRM), and tried a new approach to solving the problem of unexpected traffic behavior.
They took into account scenarios controlled by a human driver, as the driver did not favour self-driving for safety reasons.
The new artificial intelligence uses cameras and sensors to capture ambient conditions and record vehicle status data including road condition, visibility, wheel angle, weather, and speed.
The Artificial Intelligence system, which uses the « repeated neural network » (RNN), adapts to identify patterns in the data. When unique scenarios emerge that the control system cannot handle on its own, the driver will be alerted in advance to any potentially dangerous situation.
« To make vehicles more independent, many of the methods currently used are examining what cars understand about traffic, and then trying to improve the models used in them, » techXplore reported, according to The Rising Engineering.
« But the big advantage of our technology is that we totally ignore what the car is thinking. Instead, we limit ourselves to data based on what is actually happening, and look for patterns. In this way, AI detects critical situations, which the car may not be able to identify, or have not yet discovered. »
According to Steinbach, the system they developed offers « a safety function that knows when and where the weaknesses of cars are located. »
Centrally stored AI data enables entire fleets of self-driving cars to adapt, and the research team and BMW Group have tested new artificial intelligence technology with recent autonomous driving vehicles on public roads, and have tested nearly 2,500 scenarios, where the driver had to manually control after being alerted.
According to the study, AI was already able to predict scenarios that might be critical with more than 85%, even 7 seconds before the accident.
« Every time a potentially critical situation emerges in a leadership test, we make it a good example of new training, » said Christopher Cohn, one of the study’s authors. This means that large amounts of new data are needed, since AI can only identify and predict scenarios it has encountered before.
Once enough data had been collected, new data were practically generated on their own. The central storage of data allows all these scenarios to be compiled, and with each scenario of imminent death in transit being stored, the entire fleet learns how to avoid this situation in the future.