It is surprising how many of us revert to stereotypes when we get behind the wheel of a car. Suddenly, the complex emotions and thought processes which govern us the rest of the time are supplanted by base animal instincts.
Researchers at Nissan Cambridge Basic Research in Massachusetts have used the predictability of this behaviour to develop a system which is able to predict drivers’ actions with 95% accuracy.
Alex Pentland and Andrew Liu decided that the driving situation is sufficiently constrained to be able to infer drivers’ intentions from their behaviour behind the wheel. By observing the actions of drivers prior to executing a turn, lane change, brake, and so on, they were able to build up models of typical human behaviour – like the Markov models used in speech recognition technology.
In tests, sensors in a simulator recorded the hand, leg and eye movements of different subjects as they drove around a simulated world. These readings were then compared to the `Markov’ models in order to determine which action was most likely given the observed pattern of steering, acceleration and braking.
The implications for road safety are far reaching. A car which knows that you are about to change lanes will be able to warn you of vehicles driving in your blind spot. Furthermore, if external video cameras are used to monitor direction and acceleration then this kind of system could be used to intelligently control the flow of traffic.
There is no reason why this approach to modelling human behaviour could not be extended to other human-machine systems, argues Liu. In similar tasks we should be able to recognise the intended human action and build a system which adapts to better suit the user’s purpose.