Switched potential fields for navigation and control of nonholonomic and underactuated autonomous mobile robots

Traditional use of potential-field methods in mobile robot navigation has been limited to omnidirectional (free-flying) robots, and the emphasis has been on the derivation of local minima free potential-functions in the configuration-space of the system. The presence of underactuation and nonholonomic constraints in many real-life robots usually render these methods ineffective.; Nonholonomic constraints manifest in wheeled mobile robots due to no-slip constraints on the wheels, and in space robots, due to the conservation of total angular-momentum. Even if such a robot is subjected to a potential-field with a unique global minimum in configuration-space, it might still fail to reach this minimum on account of its kinematic constraints. A similar situation arises in underactuated robots, e.g., autonomous helicopters, where inputs based solely on potential-fields are not enough for system stability.; This work introduces the concept of switched potential-fields as a viable alternative for path-planning and path-following control of autonomous mobile robots with kinematic constraints and/or underactuation. In the case of underactuation, potential-fields are combined with a two-time-scale state-space decomposition for designing control-laws. The proposed method is developed and illustrated within three classes of systems. Pure nonholonomic constraints are illustrated by a planar differentially driven two-wheeled robot. Nonholonomic constraints with underactuation is depicted by an inverted wheeled-pendulum which has an unactuated non-planar degree of freedom. Application to underactuated spatial motion is exemplified by an autonomous helicopter.; The proposed method obviates the computation-time intensive step of finding a dynamically feasible time-trajectory, and its subsequent time-scaling for input-bound satisfaction. The vehicle is driven by a stabilizing control-law within an obstacle-free region, called a bubble, which is then switched to produce gross motion. The derivation of a suitable switching-law automatically guarantees speeds and inputs being restricted to specified bounds, while generating a smooth segue from bubble to bubble. The desired path can be modified in real-time based on sensor inputs. Extensions and combinations of the proposed methodology with other techniques point to its applicability for autonomous locomotion of a variety of vehicles in heterogenous media.