Abstract
We have been developing a co-worker robot which works in cooperation with workers for an automobile assembly line. The co-worker robot is not directly involved in the assembly processes but carries out some nonessential tasks for supporting workers. In this paper, we improve our old system which decides a delivery timing based on logistic regression model. We propose a method to estimate a worker's current task for improving performance of timing decision system. To achieve these, the worker's behavior is modeled combining two probabilistic techniques: Gaussian mixture model for classifying areas corresponding to each task and hidden Markov model for modeling and recognizing a worker's action. By estimating a worker's behavior using a worker's kinetic information and supporting the worker at the proper timing, this system improves work efficiency and supplies the part to the worker effectively.
