Being able to understand and carry out spoken natural instructions even in limited domains is extremely challenging
for current robots. The difficulties are multifarious, ranging from problems with speech recognizers to
difficulties with parsing disfluent speech or resolving references based on perceptual or task-based knowledge.
In this paper, we present our efforts at starting to address these problems with an integrated natural language
understanding system implemented in our DIARC architecture on a robot that can handle fairly unconstrained
spoken ungrammatical and incomplete instructions reliably in a limited domain.