Movement Primitive Sequencing via Attribute Grammars


In this work we propose attribute grammars as a mechanism to sequence movement primitives. Attribute grammars extend probabilistic context-free grammars by introducing attributes and conditions to the grammar symbols and rules. We show how such grammars can be applied to solve complex tasks by sequencing simpler subtasks. Each subtask is represented as movement primitive and the main task is solved by a sequence of primitives. By defining a general set of attributes and a corresponding evaluation scheme we introduce a general framework to transform probabilistic context-free grammars for movement primitive sequencing tasks to attribute grammars. We apply an attribute grammar to solve the task of picking and placing a stone in a game of tic-tac-toe.

ICRA 2018, Third Machine Learning in Planning and Control of Robot Motion Workshop