Universal Robots Forum

Teaching UR5 with reinforcement learning in Unity


I created a simulation model of UR5 in Unity (the game engine). I am trying to teach the UR5 to reach a target object that changes its position in space with reinforcement learning. Each component is defined as a rigid body, which means it has: mass, drag (how much air resistance affects the object when moving from forces) and angular drag.

The robot has 6 joints, each one connects two components. For each joint I attached a hinge joint. There are several ways to change the angle of a joint:

  1. Define spring and damper and change the target angle that desired.
  2. Define force (the maximum torque the motor can exert) and change the target velocity. If the velocity is zero, so there is no motion.

In addition, for each one of the 6 joints I defined limits for the angle range: from minus 180 degrees to plus 180 degrees (the same as from 0 to 360 degrees).

I could not find the mechanical and physical information needed to create good simulation that imitate the real robot. Therefore, I’m turning to you – I would like to receive such information, that can help me improve the simulation and the robot motion.

Thank you in advance,