Robot Controller Parameter Identification for Simulated Environments

Simulation of robotics systems presents many advantageous features as it can offer a safe, fast and controlled replica of real systems. Its main purpose is to test new applications or behaviours prior to deployment. It can also leverage the training of robotics systems. However, in order to achieve a reliable simulation, many physical parameters have to be appropriately chosen. Some of these parameters can be a difficult choice for the simulation designer. In this paper, we present a method to identify robot controller parameters in simulated environments. The presented methodology uses a few real-world trajectories, it iteratively minimizes the difference with virtual trajectories until convergence is achieved. The purpose of our work is not to find the optimal parameters but a suboptimal set which is realistic enough.