Learning Momentum: Integration and Experimentation

Abstract

We further study the effects of learning momentum as defined by Clark, Arkin, and Ram[1] on robots, both simulated and real, attempting to traverse obstacle fields in order to reach a goal. Integration of these results into a large-scale software architecture, MissionLab, provides the ability to exercise these algorithms in novel ways. Insight is also sought in reference to when different learning momentum strategies should be used.