Teaching motor schemes using adaptive accommodation algorithms for complex assembly

Complex assembly involves the objects with complex geometry, where unexpected contacts may occur frequently during insertion due to the complex shapes of assembled parts and the environment. In the previous study, we developed adaptive control algorithm (adaptive accommodation method) for 2D- and 3D-complex assembly tasks. In this study, we developed an artificial neural network (ANN) model for complex assembly. First, the proposed model learns assembly schemes through teaching signals from the adaptive accommodation method. Then the trained ANN model was used to execute complex assembly tasks in different geometric conditions. To test the validity of the proposed strategy, the trained ANN model trained with adaptive accommodation algorithm performed Tassembly. The simulation results showed that proposed ANN model can successfully achieve task goals as the adaptive accommodation method. For future studies, we are planning to use the developed ANN module trained by human demonstration for various contact tasks.