Workspace Modeling and Trajectory Mapping for Robotic Execution using Artificial Neural Networks

Demonstrating a task at trajectory level is one of the key concepts in robot programming by demonstration approaches. The suitability of the demonstrated trajectory for robot execution is a challenging problem to answer in such approaches. In this work, a simplified framework is presented to judge the suitability of an externally defined trajectory for robot execution. The approach is based on learning of the workspace model of the robot and then retrieving an analogous trajectory for robot execution corresponding to externally mapped trajectory. An Artificial Neural Network architecture is used to learn the workspace model of the robot and extract an analogous trajectory. The approach is demonstrated for a six degrees of freedom robot. Simulation results reveal that the approach provides an effective way to evaluate an externally demonstrated trajectory.