Many serious automobile accidents could be avoided if drivers were warned of impending crashes before they occur. Creating such warning systems by hand, however, is a difficult and time-consuming task. This paper describes three advances toward evolving neural networks with NEAT (NeuroEvolution of Augmenting Topologies) to warn about such crashes in real-world environments. First, NEAT was evaluated in a complex, dynamic simulation with other cars, where it outperformed three hand-coded strawman warning policies and generated warning levels comparable with those of an open-road warning system. Second, warning networks were trained using raw pixel data from a simulated camera. Surprisingly, NEAT was able to generate warning networks that performed similarly to those trained with higher-level input and still outperformed the baseline hand-coded warning policies. Third, the NEAT approach was evaluated in the real world using a robotic vehicle testbed. Despite noisy and ambiguous sensor data, NEAT successfully evolved warning networks using both laser rangefinders and visual sensors. The results in this paper set the stage for developing warning networks for real-world traffic, which may someday save lives in real vehicles.

No abstract

This paper describes our procedure and a software application for conducting large parameter sweep experiments in genetic and evolutionary computation research. Both procedure and software allows a researcher to examine multivariate nonlinearities that are common in genetic and evolutionary computation. Experiments of this nature are well suited to distributed computing environments (such as Grids and clusters) and we present an automated system for conducting parameter sweep experiments on heterogeneous networks. Emphasis is placed on experimental sampling, distributed robustness, and data analysis. The parameter sweep experimental procedure is easily applicable to any experiment involving computer simulations but is particularly well suited for evolutionary computation experiments.

This paper describes a tunably-difficult problem for genetic programming (GP) that probes for limits to building block mixing and assembly. The existence of such a problem can be used to garner insight into the dynamics of what happens during the course of a GP run. The results indicate that the amount of mixing is fairly low in comparison to the amount of content that could be present in an initial population.

No abstract