Fast cars could be tuned by evolution
- 16:03 18 June 2004
- Exclusive from New Scientist Print Edition. Subscribe and get 4 free issues
Can evolutionary theory help rival Formula 1 teams break Michael Schumacher's seemingly unassailable hold on top-flight motor racing? Computer scientists at University College London think it might.
Peter Bentley and Krzysztof Wloch have used genetic algorithms software that mimics evolution's drive for fitness to breed the best tuning configurations for racing cars.
Using the technique, they shaved a second off the best time achieved by an expert. They will present their results at a conference on evolutionary systems in Seattle next week.
Genetic algorithms mimic the principles of evolution to breed solutions to a problem. A population of potential solutions is tested for fitness and the best are cross-bred and mutated. The unfit members of the next generation are weeded out, simulating natural selection, leaving the fittest solutions to go on to breed.
Unfortunately Bentley and Wloch did not have any real cars to hand, so instead they applied their algorithm to virtual cars in the PC game Formula One Challenge.
This lets players set 68 variables governing the car's performance, including factors such as rev limits, gear ratios, tyre pressures and suspension damping. They say there is no reason why the same principle could not be applied trackside at Formula 1 races.
The team started with a population of randomly chosen tuning configurations, each of which was tested on two virtual tracks.
Recombination and mutation of the best 40 per cent were then used to come up with the next generation, some of which were faster still around the track. Eventually their system evolved configurations that consistently broke track records.
Bentley claims the technique would work even better if it were fed real-time performance telemetry from cars during a race. Genetic algorithms running in trackside computers could then be used to fine-tune the settings of a car during pit stops.
It could happen. UK software firm Yearstretch of Purley, Surrey, is already developing a similar technique for motor sports such as touring-car racing.
"We expect it to offer a gain in automatic set-up much as the UCL research suggests is possible," the firm's spokesman Paul Weighell says.
Genetic algorithms are already used in F1 to develop pit-stop strategies and design components. "F1 will no doubt use more GAs every year," Weighell says.
But John Nixon, a motor sport design expert at Cranfield University in the UK, warns that a car whose performance is based on evolved tuning parameters has limitations. "All of these things are based on assumptions," he says. "One crash that puts oil on a corner can throw all of them out."