Researchers at University of Michigan and MIT are devising an algorithm that models human driving behavior to help cars avoid accidents on the road. Early results of the model are expected to be published in the journal IEEE Robotics and Automation Magazine.
Mechanical engineers Domitilla Del Vecchio of MIT and Rajeev Verma at University of Michigan (formerly at MIT) are developing an approach to intelligent transportation that can better protect a car’s drivers and passengers, yet not go overboard in alerting drivers about potential scenarios that pose very low risks. They designed their algorithm to contend with other cars on the road and, if needed, take control of the vehicle to prevent a crash.
The human-behavior approach focuses on two main patterns in driving a car: braking and accelerating. Depending on the driver’s braking or accelerating mode at a given moment, there is a finite set of possible places the car could be in the future, whether a tenth of a second later or a full 10 seconds later. The algorithm takes this possible physical positions and combines it with predictive models of human behavior, for example when and where drivers slow down or speed up around an intersection.
The result is a program that can compute, for any two vehicles on the road nearing an intersection, a defined area in which two vehicles are in danger of colliding. The car with a system using the algorithm then calculates a game-theory based decision; using information from its on-board sensors as well as roadside and traffic-light sensors, the system tries to predict what the other car will do, and reacts accordingly to prevent a crash.
Del Vecchio and Verma tested their algorithm in the lab with two miniature miniature vehicles (pictured at top) on overlapping circular tracks: one autonomous and one controlled by a human driver. Eight volunteers participated, to account for differences in individual driving styles. In 97 of 100 trials, the algorithm-equipped vehicles avoided collisions, by keeping out of the defined collision area. The vehicles entered the potential collision area the other three times, and one of these instances resulted in a collision.
The engineers attribute the three failed tests to delays in communicating between the vehicles with the intelligent system and the controller workstation, which can throw off the response by a fraction of a second, long enough to allow a collision to happen. In addition to improving the electronic communications among the components, the researchers say they can make the algorithm more conservative, on the assumption that one can never completely rule out communications delays.
Next steps in the development of a real on-board system include testing on full-size vehicles with human drivers, incorporating human reaction-time data, building in sensors for weather and road conditions, and accounting for make and model-specific manufacturing details.
Read more: Computer-Vision System Unveiled for Auto Collision Avoidance
* * *
You must be logged in to post a comment.