|A Volkswagen self driving car. / Photo by: Steve Jurvetson via Wikimedia Commons|
Driverless cars cannot change lanes as efficiently as a human driver can. These vehicles depend on calculating buffer zones to avoid collisions. To optimize this calculation method, investigators at the Massachusetts Institute of Technology made an algorithm to help autonomous vehicles in better reassessing the situation on the road.
Optimization Algorithm for Driverless Vehicles
During off-peak hours, most roads have smooth traffic that helps driverless vehicles calculate other vehicles and obstacles. However, during rush hours, traffic is intense and going around other cars can be a difficult task for autonomous vehicles, but not with a human driver.
In a typical configuration, AVs calculate buffer zones around other vehicles in the area to prevent collisions. These buffer zones are also used by AVs to determine the possible next position of other vehicles. Pre-computed buffer zones enable AVs to react appropriately. Unfortunately, a traffic jam can render pre-computation of buffer zones nearly useless. This is because the use of precomputed buffer zones in a traffic jam is restrictive, causing AVs to fail in changing their lanes. So, investigators at MIT developed a new algorithm to remedy the problem.
They started with the Gaussian distribution or a common continuous probability of distribution. The use of Gaussian distribution represents the latest position of the AV and the factoring of its length and the possibility of its estimated location. Next, they used the approximations of the direction and velocity of AV to generate a logistic function. Multiplying the data from the logistic function by the Gaussian distribution provided the skewed distribution which defined the AV’s new buffer zone.
To test if the algorithm could work, they ran a simulation that included 16 AVs driving in a virtual environment. The environment had several hundreds of other vehicles that could interact with the AVs. The independent AVs were not directly communicating with each other, but have been equipped with the new algorithm.
The result of the simulation showed that the AVs did not suffer from system conflict or any collisions due to the new algorithm. Moreover, each AV had a different risk limit that resulted in various driving styles, compared to a traditional computational algorithm.
“Using the static, precomputed buffer zones would only allow for conservative driving, whereas our dynamic algorithm allows for a broader range of driving styles,” stated Alyssa Pierson, the first author of the study from the Computer Science and Artificial Intelligence Laboratory at MIT.