Sensor Adaptation andDevelopment inRobots byEntropy Maximization ofSensory Data
Chrystopher L. Nehaniv, Daniel Polani, Hatfield Herts Alio · 2005
A methodispresented foradapting thesen- sorsofa robottothestatistical structure ofitscurrent environment. Thisenables therobot tocompress incoming sensory information andtofindinformational relationships between sensors. Themethodisapplied tocreating sensori- topic mapsoftheinformational relationships ofthesensors ofa developing robot, wheretheinformational distance between sensors iscomputed using information theory and adaptive binning. Theadaptive binning methodconstantly estimates theprobability distribution ofthelatest inputs tomaximize theentropy ineachindividual sensor, while conserving thecorrelations between different sensors. Results fromsimulations androbotic experiments withvisual sensors showhowadaptive binning ofthesensory datahelps the system todiscover structure notfoundbyordinary binning. Thisenables thedeveloping perceptual system oftherobot to bemoreadapted totheparticular embodiment oftherobot andtheenvironment. IndexTerms- Ontogenetic robotics, sensory systems, en- tropy maximization