This paper introduces an hybrid system for modeling and recognition of sequences of ‘states’ in indoor robot navigation. States are broadly defined as relevant situations (in the real world) in which the robot happens to be during the navigation. The hybrid is based on parallel, state-space recurrent neural networks trained to perform a-posteriori state probability estimates of an underlying hidden Markov model (HMM) with fixed transition probabilities, given a sequence of sensory (e.g. sonar) observations. Training is accomplished in a supervised manner. Recognition is carried out either in a Dynamic Programming framework, i.e. searching the Maximum A Posteriori (MAP) joint probability of state-posteriors and transitions along paths of the HMM (useful to learn maps of the environment), or in real-time (useful for navigation itself). Encouraging experimental results of state-emission probability estimation and recognition of noisy sequences acquired by a mobile robot equipped with 16 sonars are presented
An Hybrid HMM/Recurrent Neural Networks Approach to Indoor Robot Navigation
Trentin, Edmondo;Cattoni, Roldano
1996-01-01
Abstract
This paper introduces an hybrid system for modeling and recognition of sequences of ‘states’ in indoor robot navigation. States are broadly defined as relevant situations (in the real world) in which the robot happens to be during the navigation. The hybrid is based on parallel, state-space recurrent neural networks trained to perform a-posteriori state probability estimates of an underlying hidden Markov model (HMM) with fixed transition probabilities, given a sequence of sensory (e.g. sonar) observations. Training is accomplished in a supervised manner. Recognition is carried out either in a Dynamic Programming framework, i.e. searching the Maximum A Posteriori (MAP) joint probability of state-posteriors and transitions along paths of the HMM (useful to learn maps of the environment), or in real-time (useful for navigation itself). Encouraging experimental results of state-emission probability estimation and recognition of noisy sequences acquired by a mobile robot equipped with 16 sonars are presentedI documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
