Linguistic Decision Making For Robot Route Learning

Machine learning enables the creation of a nonlinearmapping that describes robot-environment interaction, whilecomputing linguistics make the interaction transparent. In thispaper, we develop a novel application of a linguistic decision treefor a robot route learning problem by dynamically deciding therobot’s behaviour, which is decomposed into atomic actions in thecontext of a specified task.We examine the real-time performanceof training and control of a Linguistic Decision Tree, and explorethe possibility of training a machine learning model in anadaptive system without dual CPUs for parallelisation of trainingand control. A quantified evaluation approach is proposed, anda score is defined for the evaluation of a model’s robustnessregarding the quality of training data. Compared with thenon-linear system identification NARMAX model structure withoffline parameter estimation, the linguistic decision tree modelwith online LID3 learning achieves much better performance,robustness and reliability.