Present Day Robotics

Automated vision systems are tremendously useful for face recognition, finger print identification, bin picking, label placement verification on products, optical character recognition, automatic recognition of suspicious items at airports and many other applications. Although all these tasks are now being undertaken by information processing engines (the vision part of robotics), the way these machines do this is radically different from the way humans perform the same task. However these differences should not be allowed to distract us: we fly with speed and safety but the machines we use do not flap their wings. We travel over the ground in cushioned comfort but none of the machines we employ either hop, walk, trot or gallop. There are no air conditioned horses!

Let us consider a specific shop floor problem/example as encountered in the problem generally called the ‘bin picking’ problem. The task can more specifically be described as follows. A bin is being filled with some plastic parts that are coming off a conveyor being fed from a moulding machine. Our task is to program a robot to pick a part from the bin and place it in a packaging machine. The easiest solution would be to never allow the parts to get mixed up and in all sorts of orientations in the first place i.e. handle the part correctly after it comes out of the moulding machine while it is still properly oriented. This, unfortunately, is not possible for a lot of reasons (all the problems have to do with slowing the overall processes down too much) so a different solution is needed. Both mechanical orienting machines and human operators are currently used to address this problem.

What we need is a vision system that can tell the robot to pick the most appropriate part up in the same way as a human being would do. The vision system has to decide which part is to be picked up, how is it oriented, whether it is trapped under another part or not, whether the handle that we want to grasp the part with is available to the robot gripper and what to do if no suitable part can be identified. Every task in the process is difficult and we may have less than half a second to make all the decisions. The decision-making process itself can start only after the robot picks up the last part it identified or immediately after the last part that was dropped into the bin has settled down. This we cannot do with the knowhow we have at this time, however, progress is being made and solutions will be formulated. These are the concepts that have to be built upon to create a viable vision system. Work is in progress to create both an electronic retina and the electronic equivalent of a brain cell. Hopefully these can be combined by the millions to assemble a computer that will, more easily, be able to recognize objects and do useful work. Self-programming software will solve pattern recognition tasks.