A Quantitative Method for Comparing Trajectories of Mobile Robots Using Point Distribution Models

The need for efficient security systems led to the development of automatic behavioral identification tools using video tracking. However, in the field of mobile robotics, trajectories are seldom taken into account to qualify robot behavior. Most metrics rely mainly on the time to accomplish a given task or on a prior knowledge of the robot controller, with the assumption that the trajectory can be kept within a maximal bounding error. A trajectory analysis method based on a Point Distribution Model (PDM) is presented here. The applicability of this method is demonstrated on the trajectories of a real differential-drive robot, endowed with two different controllers which lead to different patterns of motion. Results demonstrate that in the space of the PDM, the difference between the two controllers can be easily quantified. This method applies equally well to the trajectories gathered in real world experiments as to those generated in a corresponding realistic simulation. Quantitative comparison between these results (real and simulated) affords an assessment of the simulation quality, when simulation features are appropriately tuned.