ROBOT TASK RECOGNITION USING DEEP CONVOLUTIONAL LONG SHORT-TERM MEMORY

The robot performs a broad range of tasks where storing and producing tasks demand an expert. The job has a succession of vectors in the created file, each needs to be analysed just using simulators, despite the simulators being computationally demanding, and an expert needs to evaluate the task. Therefore, creating a system to categorise the robot tasks is vital to enable the expert to select the task. Currently, robot specialists examine the raw robot trajectory by visual inspection using simulators based on their expertise and it takes substantial effort and time. This research offers that the one-dimensional convolutional neural network long short-term memory model (1D CNN LSTM) is used to automate the recognition of robot behaviours. Initially, the raw robot data is pre-processed and normalised. Then, one-dimensional convolutional layers extract the characteristics of robot tasks, followed by the recurrent layer retrieving the temporal information from the data. A fully connected layer produces the final output class probability representation with softmax output. The analysis of the findings revealed that the model's accuracy was 98.4% and that each class's Precision, Recall, and F1 scores were all over 97%. This study shows that robot task classification can correctly categorise tasks using robot joint trajectories.