Sparse distributed memory for experience-based robot manipulation

Sparse distributed memory (SDM) is a mathematical technique based on the properties of high-dimensional space for storing and retrieving large binary patterns. This model has been proposed for cerebellar functions, and has been used in simple visual and linguistic applications to date. This paper presents an SDM for robotic applications, especially for storing and recognising mobile manipulation actions of a 6-DOF robot arm. Sequences of events are stored as subjective experiences and are later used to guide robot arm behaviour based on its memory content. Several simple manipulation tasks, such as lift and place a wastebin from and on the floor, push an object aside on a table-top, and draw shapes in the air are analysed under different operation modes. The robot system shows good reproduction abilities of task-dependent arm trajectories based on sparse distributed memory. Moreover, the content-addressable, associative memory even predicts the residual arm trajectory of a task if the arm is placed somewhere close to a learnt trajectory.