The goal of our work is obtain a method usable to estimate in real-time muscle forces involved in the motion of the human at work. We want to use this method in order to test the ergonomics of the workstations during the conception process. This test will take place in a virtual scene in which we immerse the operator. For this reason, the method has to meet the trade-off between accuracy and performance. The application case of this method is based on the upper extremity. Our idea is that interpolation is more efficient in terms of computation time than optimization. In this article, we first present how we have built a database with the results of a muscular estimation based on an inverse dynamics approach and an optimization step. In a second time we present our interpolation algorithm, which is based on Delaunay tessellation of each joint muscle situation. For a given frame of motion capture, we search in the Delaunay tessellation that classifies the database the nearest neighbors in terms of joint position, speed and acceleration. From these nearest neighbors we perform a weighted interpolation of the muscle forces involved in the joint motion. Some results for the elbow flexion/extension joint are presented in order to discuss these results and compare with the classical approach. Further, we compare computation time since it is a fundamental point for immersion in virtual reality. At last we conclude on the perspectives opened by this new algorithm.