2023 年 31 巻 2 号 p. 60-65
A single spherical synchronous actuator can drive multiple degrees of freedom. This simplifies the mechanism and is expected to be applied to industrial robots. Position control of the actuator is performed using torque maps, which are torque data in each posture. However, measurement errors in the torque map reduce control accuracy. To solve this problem, a new 2-DOF control method using a deep reinforcement learning algorithm was proposed. However, that method could not control 3-DOFs because of the local optimum solution. Therefore, we extended this method to 3-DOF and investigated a new reward function that can be driven with high accuracy. As a result, 3-DOFs could be driven with an average angular error of about 0.4 degrees.
