Autonomous car: a new simulation tool to differentiate driver behavior

This new technique presented in detail in the review IEEE Robotics and Automation Letters is able to detect and classify the driving behavior of other vehicles and to differentiate between drivers with “aggressive” behavior and those with a calmer and more reassuring driving style.

The problem with self-driving cars

As its name suggests, an autonomous car is a vehicle designed to be able to drive autonomously without the intervention of a driver. This type of vehicle should considerably reduce stress during journeys since the driver will no longer have to worry, for example, about the clutch or managing the speed of the vehicle.

>> Read also: Autonomous cars will communicate with each other

This is not the only advantage of an autonomous vehicle. Indeed, autonomous driving should also make it possible to achieve fuel savings thanks to better route management, easier parking, particularly in built-up areas and improved autonomy for people with disabilities.

But before being really usable on all road networks, the self-driving cars still require a number of adjustments. The technology is not yet perfect, and before being deployed on a large scale, autonomous vehicles must be able to move on a wide variety of road networks and in all driving environments. In addition, these vehicles must be able to anticipate driving in the event of difficult weather conditions, when an obstacle suddenly appears on the road or even when another type of vehicle commits carelessness or a driving error.

It is precisely in an attempt to improve the behavior of autonomous cars in the face of other drivers that American scientists and engineers have developed a new technique which could well improve the efficiency of driving simulators used to train navigation systems. autonomous vehicles.

A new simulator focused on driving behavior

Scientists at the University of Maryland have created a new kind of simulator that analyzes the driving behavior of drivers in order to allow autonomous vehicles to adapt their trajectories and their safety measures accordingly.

© metamorworks/Shutterstock

Artificial intelligence at the service of the autonomous car.

Indeed, it is possible to classify the behavior of drivers into two main categories: drivers with an aggressive and sometimes belligerent tendency and drivers who are calmer and more attentive.

By detecting these different driving behaviors, the researchers succeeded in developing a simulator that takes into account the behavior of other human drivers. It is then able to interpret and recreate these different behaviors found in real traffic situations.

Using artificial intelligence, researchers have developed a model-based simulation technique capable of classifying the driving behavior of human drivers and possibly other autonomous vehicles. This model, called CMetric, works by analyzing and calculating the trajectories of other vehicles using, among other things, computer-managed vision tools.

In addition, in order to improve the navigation of autonomous cars, the researchers introduced a method for predicting the behavior of other vehicle drivers through complex algorithms based on the Deep Q-Network (DQN). This driving behavior prediction model can be integrated into many algorithms for vehicle navigation.

>> Read also: Electric, autonomous, connected… what will your car look like in 10 years?

Improving the performance of simulators in urban environments

Until now, the majority of simulators intended for the training of autonomous vehicles have had great difficulty when it comes to navigating in complex urban centers, especially on roads with very dense traffic and equipped with many lights. signaling. The difficulty increases further when pedestrians and bicycle users are added to the simulation.

While current systems of autonomous driving are primarily intended for traffic on freeway networks, the simulation technique developed by researchers at the University of Maryland represents a major advance intended to improve the performance of simulators intended for learning autonomous vehicles in urban centers .

This simulation technique is above all a tool intended for training navigation algorithms for autonomous vehicles in complex urban environments. The researchers have also decided to make this simulation technique open source and therefore freely accessible to car manufacturers.

This CMetric model is therefore a revolutionary development which could eventually replace the old simulation platforms which do not take into account the behavior and the way of driving of other drivers.

This new system is also interesting for manufacturers of conventional (non-autonomous) vehicles, which will be able to use this technology to improve driver assistance systems.

>> Read also: Autonomous car: a researcher discovers a risk of a giant traffic jam

Source: Angelos Mavrogiannis, Rohan Chandra, Dinesh Manocha, “B-GAP: Behavior-Rich Simulation and Navigation for Autonomous Driving”, IEEE Robotics and Automation LettersVolume: 7, Issue: 2, April 2022,

We wish to say thanks to the writer of this article for this awesome web content

Autonomous car: a new simulation tool to differentiate driver behavior


We have our social media profiles here , as well as other pages on related topics here.https://www.ai-magazine.com/related-pages/