A Variable Prediction Horizon MPC Approach for Leader-Follower Transportation in Presence of Obstacles

Cooperative transportation using a Multi-Robot System has emerged as a relevant research topic in recent years, primarily due to its potential in challenging workspaces, including transporting large payloads in cluttered environments. In this case, an obstacle avoidance feature is essential to guarantee the safe navigation of both agents and payload. This thesis addresses the problem of cooperative transportation in environments populated by static and dynamic obstacles by implementing a feasibility-aware leader-follower Model Predictive Control based algorithm. Compared to the existing state-of-art, a more realistic model of agents dynamics is considered. A geometrical approach is employed to define the hard constraints for obstacle avoidance, while soft constraints are represented by a potential repulsive field functional cost, which acts to guide the agents away from the obstacles. The results achieved through numerical simulations in MATLAB show that a variable prediction horizon MPC can guarantee improved performance in trajectory planning with a reduced computational effort, compared to a fixed prediction horizon MPC. Algorithm robustness to disturbance and delay is also assessed, showing that it depends both on MPC parameters and obstacle placement. Lastly, Gazebo simulations are performed to include physics and implement more realistic experiments, showing that this algorithm is able to fulfill its tasks successfully.