Line-of-Sight Control of a Spherical Parallel Manipulator via Deep Reinforcement Learning

In this paper, controllers solving a rest-to-rest line-of-sight orientation problem are developed in a deep reinforcement learning framework and applied to a particular spherical parallel manipulator. Perfect kinematic information is assumed for the development and training of the neural networks. Extensive Monte Carlo simulations are performed to test the novel controllers in noisy environments and to compare their performances with an efficient unconstrained feedback controller. Results from an experiment with sole joint position readings are shown to validate the proposed control approach.