Giving robots a sense of touch
Practice makes perfect (controllers) - by Raul Cruz-Oliver
Robotic manipulators are a well-established tool in industry. Usually, these machines allow control only via deterministic open loop position or velocity commands due to safety reasons. Although this is enough for many applications, it is not suitable when there is contact with the environment, since the actual position of the robot is always subject to errors. To deal with this, force feedback control has been introduced. It consists in giving a “sense of touch” to the robot, so it “knows” how much force it is applying to its environment and can adjust it accordingly.
The most classic approach for force control presented in the literature is commanding torque to the robot actuators given the measured force. However, the actuator torque interface is usually not directly accessible; only position and velocity can be commanded.
We address this limitation by using an outer control loop that interacts with the velocity interface. The controller uses information from an external force sensor mounted in the robot’s end-effector (red in the picture) to command the joint velocities in a way to achieve a robot behaviour that resembles traditional force control, the above mentioned “sense of touch”.
This outer control loop has a large number of parameters that are extremely hard to tune and depend on the robot itself, but also on the specific task that the robot is performing. To find an auto-tuning procedure for this family of controllers, our researchers explore stochastic optimization methods like Iterative Learning Control (ILC) to perfect the controller performance from iteration to iteration.
