On the Impact of Gravity Compensation on Reinforcement Learning in Goal-Reaching Tasks for Robotic Manipulators

Advances in machine learning technologies recent years have facilitated developments autonomous robotic systems. Designing these systems typically requires manually specified models of the system and world when using classical control-based strategies, or time consuming computationally expensive data-driven training learning-based strategies. Combination control strategies may mitigate both requirements. However, performance combined is not obvious given that there are two separate controllers. This paper focuses on one such combination, which uses gravity-compensation together with reinforcement (RL). We present a study effects gravity compensation algorithms solving reaching tasks simulated seven-degree-of-freedom arm. The results our demonstrate coupled RL can reduce required involving elevated target locations, but all locations.