Describir: Reinforcement learningApproach to Objectcontact Motion with Estimation of Lowdimensional Submanifold and Mode-Boundary