Combining Stochastic Automata and Classification Techniques for Supervision and Safe Orchard Navigation

Abstract

Cost drivers in commercial orchards are time-consuming tasks as the drive through rows for spraying, cutting grass or collecting fruit. An automated tractor can be an answer to enhance production efficiency. For this to be acceptable by public and authorities, safety and reliability are crucial, hence information redundancy is needed to achieve a fault tolerant system. This paper addresses ways to extract information from laser scanner data. A Gaussian Mixture model is used to classify laser data into obstacles, while through diagnosis, a stochastic automaton model gives a semantic position estimate relying only on laser perception. Results demonstrate the feasibility of implementation in an autonomous tractor that use diagnosis and active fault-tolerant control to enhance availability and safety.