Abstract

At first glance, the modelling of a robot arm might seem an easy task. Indeed, much effort in this area seems to involve the machine-vision problem of determining the position of the arm with relation to the object of our interest. Once this problem of ‘target location’ has been solved, the problem remains of moving the arm to that target position. This ‘secondary’ problem of arm control will be examined in this paper.

The problem is traditionally thought of as a real-time programming problem, involving the coordination of a set of concurrent activities that have to satisfy strict timing and space constraints. There are two ways of approaching the problem, either to use an ‘ad-hoc’ method, or to formulate a more structured result. The real-time systems approach is to formalise the system, resulting in a ‘generalised’ solution akin to a real-time scheduler.

EM offers us the opportunity to explore the control of a robot arm in the ‘ad-hoc’ fashion, allowing various techniques to be tested in various situations, and to evaluate and compare their effectiveness. Taking an ‘informal’ approach to solving this problem allows a broader set of solutions to be considered.

For instance, if we consider each distinguishable position of the arm to be a node in a neural network, connected to the nodes representing each adjacent position, the task becomes one of energy minimisation, where the ‘energy’ of the arm is related to its distance from the object position.

Another approach is to move the arm in a similar way to a human arm, an ‘algorithm’ that has developed over thousands of years, and other similar methods. The important factors then become the prioritisation of the various degrees of freedom (do we rotate the whole arm, straighten the forearm or lower the wrist to pick up something just below the hand?).

The ability of the various algorithms to deal with restrictions on the robot arm is also explored. For instance, if the joints are limited to those movements possible by a human arm, or if objects are placed between the arm and the target, blocking its usual path.