Force feedback

Citation

Loughlin, C. (2007), "Force feedback", Industrial Robot, Vol. 34 No. 3. https://doi.org/10.1108/ir.2007.04934caa.001

Force feedback

Most robots use electric motors and gears to move their limbs, but the very gearing that enables the motors to move large masses also means that the motors themselves lose pretty much all sense of feeling regarding what they are doing. Basically robot control comes down to a plan for moving from A to B and keeping, your fingers crossed that nothing gets in the way.

In a fully structured industrial environment where everything is exactly where it should be then everything is fine. The arm does not need to know if it hits anything, because it would not. As soon as you move a robot into an unstructured environment then the rules change and sensing collisions becomes high on the agenda.

People have taken various approaches to tackling this problem. The easiest is a 6-DoF strain sensor mounted between the robot's wrist and the end-effector. This works up to a point provided it is the tool that first comes into contact with the obstacle. However, if the collision is further back along the arm, and remember that about 95 per cent of the robot is behind the wrist, then the wrist sensor will not detect it.

Another approach is to cover the external surface of the arm either with a tactile skin or alternatively with an array of distance sensors. The tactile skin will typically only be a few millimeters thin and, therefore, of limited use in actually preventing or minimizing collisions. By the time you have sensed the collision it is already too late and something has got bent. The array of distance sensors is more promising but these are expensive and depending on the type either have a minimum sensing distance of a few centimeters (ultrasonic), or vary in sensitivity with the type of material they are about to contact (capacitive). They work – but only in a “better than nothing” sort of way, and usually at the expense of restricted movement and of course increased system complexity.

A while ago I was working on a controller for a hydraulic manipulator and I thought I would try to create a collision detector by monitoring the difference between where I had told the robot to go and where it actually was. The idea was that if it did collide with something then the arm would be stopped from reaching its target and this would make the alarm bells ring in my software.

No such luck – what actually happened was one of two things – either the object it collided with got pushed aside or the robot lifted itself and its platform into the air.

In my view, the only practical ways to detect that an arm has come into contact with something is if you either have strain gauges on each drive shaft or you are monitoring the motor's current draw. In both cases you need to make the sensing and integral part of the servo control system rather than as a software “fuse” that only clicks into play when certain limits are exceeded.

We have managed to get away without this sort of sensing because the areas of application of robots have been quite limited. If we want to work alongside robots and have them working alongside us in service and healthcare applications then we need to take force sensing very seriously indeed. If we do then we will find all sorts of new applications will become both more accessible and less complex.

Clive Loughlin