Hybrid Model Predictive Control for Autonomous Emergency Evacuation Robot

This paper presents an approach for optimizing trajectories of autonomous vehicles in evacuation scenarios, particularly for the transportation of handicapped individuals, animals, and products. First, the hybrid model predictive control (HMPC) technique takes into account partially known infrastructure and dynamic obstacles, and allows for multiple secure destinations. The approach utilizes a platform-specific dynamics model, a global cost-to-go (CTG) function, and contextual awareness. In addition, the CTG function is generated using a Dijkstra algorithm and pseudo priority queues (PPQ), and allows for efficient trajectory planning on large maps. Furthermore, the control actions are generated in real-time using a high-frequency MPC process based on stochastic dynamic programming (SDP), using the most recent CTG information and a model of the vehicle dynamics and the perceived environment, including static and dynamic objects and skid-slip effects. Finally, the effectiveness of the proposed technique is validated through simulations.