[1I3-J-2] Machine learning: advances in reinforcement learning

[1I3-J-2-02] Generalized goal oriented deep reinforcement learning for robot arm training with continuous action space

〇Kimaura Tomoaki², Masaya Watabe², Katsuyoshi Sakamoto¹, Kouichi Yamaguchi¹, Dinesh Malla³, Tomah Sogabe^1,3,4 (1. Advanced science and technology department, The University of Electro-Communications, 2. Mechanical department The University of Electro-Communications, 3. Grid Inc., 4. i-perc, The University of Electro-Communications)

Keywords:Robot arm, Deep reinforcement learning, Continuous action, Generalized goal

In multigoal reinforcement learning, Universal Value Function Approximators(UVFA) that takes not only a state but also a goal for inputs is used. We designed a task by bringing the end effector of the 7DOF robot arm to the goals using UVFA based multigoal reinforcement learning, Meanwhile, we performed the equivalent task by changing the number of goals. We confirmed a superb prediction ability by mapping the goal reachability degree using UVFA.