Hybrid model predictive control for the emergency evacuation robot under the presence of disturbances and dynamic obstacles

Autonomous vehicles can be used for evacuations, particularly for assisting disabled people; they can also be used for moving animals and goods. In those cases, the platform must deal with partially known infrastructure and other context characteristics, including dynamic obstacles. The destination may not be unique, as multiple safe destinations may be feasible. This paper proposes a hybrid model predictive control (HMPC) approach for performing trajectory optimization. The approach exploits the available global cost-to-go (CTG) function and a model of the platform dynamics, in addition to awareness information of the nearby context. The CTG function implicitly allows the selection from multiple destinations in the context of operation. First, the Global 2D CTG is generated and periodically updated at low frequency using, in this case, pseudo priority queues (PPQ) based Dijkstra algorithm, which can generate optimal CTG functions of large maps, requiring low processing cost. Secondly, an MPC process based on stochastic dynamic programming (SDP), running at high frequency, generates the actual control actions, in real-time, based on the last available CTG information, in combination with a model of the platform dynamics and the perceived description of the surrounding area, including static, dynamic objects, skid-slip effects. Finally, simulations are performed to validate the proposed planning and control scheme.