Neural Network Task Time Mapping of a 3-DOF Clavel’s Delta Robot

The throughput of robotized pick and place cells can be improved by scheduling algorithms that determine optimal task sequences using estimates of the robot’s cycle times; the latter however depend on factors such as the robot’s dynamics, the motion planning algorithm, and the specific configuration of the task, which is often not known in advance. Exact evaluation of the task times is therefore a computationally expensive procedure that should be performed as sparingly as possible during real time calculations. In this paper a recently developed architecture for the efficient approximation of a manipulator’s task
times is applied to a 3-DOF Clavel’s Delta robot. Such an architecture is composed of a dynamic and kinematic model of the robot, of an optimizing motion planner, and of a neural network able predict the task
times quickly enough to be used in online process-optimizing scheduling algorithms.