Soft continuum robots undergoes nonlinear spatial deformation. Deep reinforcement learning (DRL) suffers from large training dataset and high time consumption. This paper reports a generalized principle, N-space (NS) framework, that employs on value-based reinforcement learning algorithms (e.g. DQN and DDQN), to overcome the challenge of local minima associated with sample efficiency or limited time consumption in online training. The performance of NS-augmented DRL (NS-DRL) was examined on controlling a self-design rope-driven soft continuum robot with 5-degrees of freedom (DoF). In this framework, the action space of the robot was divided to 6 sub-action spaces. Subsequently, the target position was divided into six sub-target positions, determined by the volumetric vector projection in each sub-action space. The action sequence was determined by actions of vector projection of the target on each sub-action space. NS-DDQN increased the convergence speed by more than 100-fold, from over 100,000 steps to approximately a thousand steps, and reduced the positioning error by over 10-fold, from over 20 mm to less than 1 mm, compared with non-NS enabled DRL positioning. The performance augmentation was also tested in DQN, implying the generalization of NS strategy in controlling soft continuum robots.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.