Design of the Perceived Safety Toolkit Sample Clauses

Design of the Perceived Safety Toolkit. For the prediction model of the PST we used a design based on neural networks. Neural networks learn (or are trained) by processing examples, each of which contains a known “input” (or combination of “inputs”) and a “result”, forming probability-weighted associations between the two, which are stored within the data structure of the net itself. The neural network is constructed from three types of layers: input layer, hidden layers, and output layer. As we want to design our prediction model based on the datasets, we have gathered during both experiments presented in Deliverable 1.4 and Deliverable 2.4, the machine learning based on neural networks presents a suitable solution. From the data from both experiments, we have created a training data set. In the dataset, every input has five parameters. In total, we had 84 different sets of inputs, and for each input, we had 30 sets of outputs. Altogether we used 2520 different sets to train our model. Every output had four different parameters. Part of the data set is shown in Table 1. For each parameter in the dataset, we have defined a numerical value used in the dataset. In Table 2, the defined numerical values for movement types are presented. Numerical values for tool types (safe tool, dangerous tool, combination of both) are stated in Table 3. Distance between the user and the robot was categorised into four categories; numerical values are presented in Table 4. Human-robot interaction was categorised as coexistence or collaboration. In Table 5 numerical values for the type of HRI are shown.