Context-Sensitive Grammar-Based Planning: A Methodology for Interpreting Task-Oriented(Natural-Language-Like) Robotic Instructions

摘要

Recently, several research efforts have been devoted to the development of natural-language-like man-machine interface technologies in the field of industrial robotics. These efforts attempt to develop interfaces that can accept high-level task specifications that resemble what people usually conceive when defining a manipulation task. The goal of these efforts is to ultimately release the operator from directly working with the low-level tedious machine codes, and at the same time, enable him/her to become an effective strategic planner in problem-solving. As motivated by this trend, our present work investigates an automatic means of synthesizing a low-level robot-oriented control program from a task-oriented, natural-language-like description. Such a description is concerned only with the state transitions in the physical world of the robots, as brought about by a specific robot operation. Specifically, we introduce a formal grammar-based methodology for task planning that derives a detailed action plan by recursively replace the right-hand side of a grammatical production rule with another rule, until an action-routine-level program is reached; this is analogous to the derivation of a sequence of terminals in a natural language grammar.