Application of multiobjective genetic programming to the design of robot failure recognition systems

We present an evolutionary approach using multiobjective genetic programming (MOGP) to derive optimal feature extraction preprocessing stages for robot failure detection. This data-driven machine learning method is compared both with conventional (nonevolutionary) classifiers and a set of domain-dependent feature extraction methods. We conclude MOGP is an effective and practical design method for failure recognition systems with enhanced recognition accuracy over conventional classifiers, independent of domain knowledge.