Although self-driving cars are often considered as superior to the human being, there are still some things these machines can learn from us, such as changing lanes. Most of the lane-change algorithms are based on detailed statistical models, which are too complex to make a decision in the thick of traffic. Others force the car not to change the lane at all, which is very impractical. However, MIT might have a solution to this problem. Researchers at MIT’s CSAIL developed an algorithm that lets autonomous cars changes lines more like humans do.
This new algorithm is basically the modification of the concept of “buffer zones” that were already used in self-driving car development. the zones determine the space around the vehicles, which is necessary for autonomous cars to avoid a collision. Earlier system calculated the space in advance to save time, which can still take too long in a fast-moving traffic.
To solve this problem, MIT uses a “mathematically efficient” approach, which lets the cars continuously adjust buffer zones on the fly. The process relies on fewer equation variables and can be customized based on the aggression level. However, there’s always a safety guarantee, which should save you from having to skew.
The project is backed by Toyota, which is also funding MIT to develop an alternative navigation system named MapLite. The system lets self-driving cars navigate country roads, which aren’t well mapped.