Intelligent navigation in space under supervisory control

Although the problem of obstacle avoidance in intelligent mobile systems has been investigated and reported in the literature, the following issues have not been adequately addressed: Intelligent navigation in unknown terrain, where the map is not available and the system is heavily dependent on sensory information for obstacle recognition and consequent path planning Inadequacy to accommodate the information on vehicular dynamics in the algorithm Computational efficiency and robustness This paper presents a discrete-event language-theoretic supervisor to implement the decision process of intelligent navigation. The specific problem investigated is navigation of an autonomous vehicle in an unmapped environment with robust obstacle avoidance and minimal computational overhead. The proposed supervisor is computationally simple and is designed to be executed in real time at a reduced frequency. The data interpretation module of the supervisor makes use of the continuous sensory information over a predetermined time slot to produce a discrete dynamic model of the current navigation scenario. Sensory information consists of obstacle data in the vehicle surroundings and external navigation commands. The control architecture optimally integrates multiple data streams and creates a hybrid language theoretic representation of the instantaneous situation. The model incorporates the information on relative positions of the obstacles and modifies the weight on each possible directions of vehicle motion depending on obstacle data, vehicle dynamics and navigation commands. The navigation supervisor generates a symbol as the optimum decision which is then interpreted as a trajectory fragment. This supervisor is well-suited for local mapless navigation since a local obstacle map is constructed online in the form of a discrete dynamic model, where no prior knowledge is necessary. However, the algorithm does require an identified model of vehicle dynamics that can be constructed off-line through well-established system identification techniques. The simulation problems studied in the paper are motion of a simulated vehicle through a static unmodeled debris field.. The algorithm is being verified on a Segway robotic mobility platform in the Networked Robotic Systems laboratory of Penn State.