TOWARDS TEMPORAL INFERENCE FOR SHAPE RECOGNITION FROM WHISKERS

 Previous  work  on  shape  recognition  from  whisker-like

sensors  has  taken  two  approaches.  First,  static  analysis from  beam  theory  has  been  used  to  estimate  contact points.  Second,  dynamic  analysis  has  estimated  contact points by considering vibrations set up on contact. This paper  proposes  additional  methods  which  utilize  full time-series  data  from  whiskers.  Full  generative  models may  be  intractable  but  approximated  by  discriminative classifiers.  Early  results  show  that  object  location  and velocity may be discriminated simultaneously.