Model-Based Chassis Control of a Wheeled Mobile Robot on Soft Ground Using the Example of the ExoMars Planetary Exploration Rover

Abstract

The exploration of the universe has greatly advanced by the usage of planetary exploration rovers in the last two decades. Major findings were only possible due to the extended range of scientific instruments onboard rovers. NASA's Mars Exploration Rover Opportunity has only stopped operating last year, after 14 years, and thereby exceeded its planned mission duration by a factor of 56. The locomotion system, consisting of single wheel drives and steering actuators as well as passive kinematics, has thus proven to be robust and reliable. Despite this success, the problem of traversing soft, sandy terrain has been ubiquitous for the NASA rovers. Opportunity's sibling, Spirit, got irretrievably stuck in a soft sand field after breaking through the thin crust that covered it. Opportunity itself almost encountered the same fate when it had wheel slippage of over 98% for about one month. During the operations of their still active Curiosity rover, the NASA/JPL team decided to avoid certain sand fields in the future. All mentioned and planned rovers feature an over-actuation due to simplicity of construction and redundancy reasons, i.e. they have more actuators than they would need for realizing a desired movement. The resulting degrees of freedom are, however, not used for optimizing the locomotion but eliminated through geometric constraints in current missions. This thesis therefore addresses the question whether a chassis controller can be designed for the control allocation of this system, such that the degrees of freedom are exploited for improving the rover's locomotion in soft sand. Although this work is derived with the example of the Rosalind Franklin rover of the European ExoMars mission, it is transferable to the field of mobile robots on rough terrain in general. The approach of this thesis is a comprising model development of the robotic rover system and the wheel-ground interaction for soft sand, called terramechanics. A fully modelbased chassis controller is then synthesized, which, at the same time, allows for explicit computation of the actuator trajectories and exploitation of the full solution space of the over-actuation. For managing these conflicting goals, the full model is divided into dynamic but determined sub-systems and a linear, static, underdetermined system of equations with several necessary model modifications. The nonlinear control methods feedback linearization and dynamic extension are used for the dynamic sub-systems, whereas the general solution to the underdetermined system of equations is used for the allocation itself. It consists of the least-squares solution with a pseudo-inverse of the allocation matrix and the homogeneous solution with a base of its null-space. The performance of the designed chassis controller is evaluated in a co-simulation with a high-fidelity model of the rover and the robustness of the approach is investigated by the injection of different disturbances.