Model-based Control for Reconfigurable Mobile Robots

We propose to develop a model-based control framework for wheeled mobile robots that robustly actuates a mechatronic robot drive with several steered wheels. Basis for our controller is a drive model that details the geometric wheel alignement and the dynamic behavior of the wheel`s steering and angular-speed actuation for the operational and some fault conditions. Our proposed control framework uses this model to solve two major tasks through on-line reasoning methods that build upon theoretical kinematics, systems-analysis, filtering- and diagnosis techniques, and control-theory. The first task provides the current mode of operation or failure for every individual wheel through state-estimation based on measurements and model-predictions for possible mode evolutions. The second task uses the estimated mode of operation or failure, derives the kinematic constraints for this mode and deduces the coordinated actuation for the rotational speed and steering angle for each individual wheel to obtain the desired movement of the mobile robot. This on-line utilization of the robot drive`s model enables our proposed control system to autonomously adapt itself in order to overcome fault conditions such as the loss of steering/actuation or traction of individual wheels but equally well handles changed operational conditions such as retracted wheels or a dynamically changed drive geometry.

Wheeled mobile robots are built to accomplish specific transport or mobility requirements. As a consequence, one obtains a large variety of robot drives with diverse functionalities and mobility characteristics and requires drive- specific controllers to operate them according to their kinematics. This project pursues a radically different approach. Instead of building specialized controllers, we propose to build a generic controller that can operate almost any robot drive. Moreover, we built a self-aware controller that analyzes a robot`s drive in terms of its mobility characteristics and automatically derives the appropriate control strategy. It performs this operation on the basis of a model that specifies the geometry and functionality (steered-, un-steered wheel, etc.) of a drive. Our intention is not just to drive every robot, but also to provide this capability robustly so that faults in the robot can be handled automatically as well. As a consequence, we allow the model to change over time in order to reflect the current mode of operation or failure of the robot and derive the controller on demand during run-time of the robot. To achieve this functionality we developed the theory and computationally efficient algorithms for two key components: (a) an estimation- or diagnosis unit that identifies the health state of the robot drive and (b) a kinematics reasoning unit that analyzes the mobility capabilities of a drive at its mode of operation or failure (as proposed by the diagnosis unit). To demonstrate this functionality we further developed a novel modular robot system for which we obtained patent protection. The modular robot system uses the nature-inspired honey-comb geometry as its basic shape for the individual robot components. In that way we obtain a versatile robot kit that enables us to build various robot drives and robots that reconfigure themselves during operation. One emphasis of this project was to keep the set of possible robot-drives as open as possible. As a consequence, we ended up with a control scheme that allows us to operate (a) almost any robot, (b) robots that change their shape/geometry and (c) any collection of robots that perform coordinated movements. Many interesting applications in factory logistics and multi-robot object transport can thus be solved easily.