MIT has developed a new algorithm for self-driving cars to make them act more like humans. Self-driving cars are notoriously bad at changing lanes, so researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) decided the answer would be to make them act more like human drivers.
Most current lane-change algorithms are based on detailed statistical models that force the car to act very conservatively - in many cases avoiding changing lanes all together. The new MIT algorithm actually gives the cars less information allowing them to make decisions faster and allows them to act more aggressively.
Less information speeds up decision making
Although this seems counterintuitive for a self-driving car it is actually necessary to help the cars learn how to navigate dense traffic conditions like New York City. "The motivation is, 'What can we do with as little information as possible?'" Alyssa Pierson, a postdoc at CSAIL and first author on a paper detailing the algorithm, said in a statement. "How can we have an autonomous vehicle behave as a human driver might behave?"
The new algorithm works by adjusting allowable buffer zones around the car. Giving the car less data allows the car to adjust this zone on the fly, and gives permission to the car to drive in different styles such as aggressive or conservative.
Lab testing so far proved successful
“The optimization solution will ensure navigation with lane changes that can model an entire range of driving styles, from conservative to aggressive, with safety guarantees,” says Rus, who is the director of CSAIL.
MIT said the new algorithm has been tested extensively in simulation in the lab but that no real-world test have yet to be conducted. The algorithm was tested in a simulation that had up to 16 autonomous cars driving in an environment with several hundred other vehicles.
Pierson described the testing, saying, “The autonomous vehicles were not in direct communication but ran the proposed algorithm in parallel without conflict or collisions. Each car used a different risk threshold that produced a different driving style, allowing us to create conservative and aggressive drivers. Using the static, precomputed buffer zones would only allow for conservative driving, whereas our dynamic algorithm allows for a broader range of driving styles.”
The research is being done in conjunction with a project called MapLite that is aiming to allow autonomous cars to drive with very little road information. MapLite proposes that rather than giving self-driving cars highly detailed maps that are difficult to create and maintain, simplified maps and special sensors will get the same results. Both projects are partially funded by the Toyota Research Institute.