Data-driven Robotic Automation using Trust-based Kalman Consensus Filter for Collaborative Assembly

Robotic automation of complex and highly dexterous manipulation tasks often make use of accurate modeling techniques, such as Denavit-Hartenberg (DH) parameter model, to ensure the high precision needed for the task. Such model-based approach, however, renders the overall manipulation slower and more expensive due to the unmodeled dynamics. Meanwhile, in multi-agent systems, it is impractical that every agent is supposed to be all equal in trust. In this work, we present a trust-based multi-robot precision modeling and filtering technique. By using the Dynamic Linearization Technique (DLT), the system parameters are estimated in real time. Trust mechanism is designed with the self-trust updating and mutual-trust updating. A data-driven Kalman consensus filter is used with the Fisher information matrix to build another trust mechanism for consensus estimation in discrete time. The proposed approach not only improves the automation throughput of complex manipulation tasks at a reduced cost and high reliability but also offers a holistic manufacturing framework enabling rapid optimization of the product and production process. Simulations presented in the work confirm the validity of the trust-based data-driven modeling and filtering methods for collaborative assembly.