Payload Parameters Identification Using Incremental Ensemble Learning

This paper introduces a new method for identifying the payload inertial parameters for collaborative robots (cobots) that operate in dynamic manufacturing environments with frequent reconfiguration. The proposed method eliminates the need for specific excitation trajectories during the identification process, allowing the identification of payload parameters while the cobot follows any arbitrary task trajectory. This is achieved via an incremental ensemble model, which utilizes incremental neural networks as weak learners. The proposed method successfully adapts the ensemble model to new task paths while maintaining accurate estimations for the payload parameters. The mean absolute error for mass, center of mass, and the moment of inertia parameters were 0.01 kg, 0.0068 kg.m, and 0.0008 kg.m(2), respectively. These results demonstrate the effectiveness of the model in dealing with new task trajectories in comparison to traditional or batch ensemble models.