New algorithm allows self-driving cars to change lanes better

Researchers at MIT have developed an algorithm that allows self-driving cars to change lanes more like people. Until now, self-driving cars were still mostly based on static information.

The researchers’ new algorithm should enable self-driving cars to change lanes more frequently, faster and more aggressively, making the decision to change lanes based on readily available, up-to-date information. The algorithm concerns the minimal information that a self-driving car needs to change lanes in a human way, as it were.

According to the researchers, most existing algorithms are not sufficient, because they have two drawbacks. For example, there are algorithms that are entirely based on detailed static models of the driving environment. These are difficult to analyze spontaneously and directly. There are also models that are so simple that it only leads to conservative, cautious, impractical decisions, often resulting in the decision not to change jobs after all.

The new model of the MIT researchers is based on buffer zones around the vehicle. This technique is already used in existing algorithms, but according to the scientists this is limited to pre-calculating the zones. The MIT algorithm also uses this technique, but if it leads to inhumane decisions, new buffer zones are calculated in real time. Especially in heavy traffic, according to the researchers, only the use of pre-calculated zones is too restrictive.

The algorithm was tested in a simulation with sixteen autonomously driving cars. Each car was set to a different risk profile, resulting in different driving styles. The cars all drove side by side with no conflict or collision. The research will be presented at the International Conference on Robotics and Automation.