Crane collision modelling using a neural network approach

The objective of the present work is to find a Collision Detection algorithm to be used in the Virtual Reality crane simulator (UVSim ®), developed by the Robotics Institute of the University of Valencia for the Port of Valencia. The method is applicable to boxshaped objects and is based on the relationship between the colliding object positions and their impact points. The tool chosen to solve the problem is a neural network, the Multilayer Perceptron (MLP), which adapts to the characteristics of the problem, namely, non-linearity, a large amount of data, and no a priori knowledge. The results achieved by the neural network are very satisfactory for the case of boxshaped objects. Furthermore, the computational burden is independent from the object positions and how the surfaces are modelled; hence, it is suitable for the real-time requirements of the application and outperforms the computational burden of other classical methods. The model proposed is currently being used and validated in the UVSim Gantry Crane simulator.