Simulated road following using neuroevolution

This paper describes a methodology wherein genetic algorithms were used to evolve neural network controllers for application in automatic road driving. The simulated controllers were capable of dynamically varying the mixture of colour components in the input image to ensure the ability to perform well across the entire range of possible environments. During the evolution phase, they were evaluated in a set of environments carefully designed to encourage the development of flexible and general-purpose solutions. Successfully evolved controllers were capable of navigating simulated roads across challenging test environments, each with different geometric and colour distribution properties. These controllers proved to be more robust and adaptable compared to the previous work done using this evolutionary approach. This was due to their improved dynamic colour perception capabilities, as they were now able to demonstrate feature extraction in three (red, green and blue) colour channels.

The final publication is available at Springer via http://dx.doi.org/10.1007/978-3-319-18084-7_2

Paper published as: Narayan A., Tuci E., Labrosse F. (2015) Simulated Road Following Using Neuroevolution. In: Headleand C., Teahan W., Ap Cenydd L. (eds) Artificial Life and Intelligent Agents. ALIA 2014. Communications in Computer and Information Science, vol 519. Springer, Cham