Examinator: Knut Åkesson, Inst för elektroteknik
This thesis proposes a Nonlinear Model Predictive Control (NMPC) approach to drive and carry objects between two Autonomous Transport Robots (ATRs) in a factory environment. Considering static obstacles and a pre-defined node network, a rough global path to goal is generated using visibility graphs and the A* algorithm. This path is used as a reference path when calculating a trajectory for one or two ATRs. The trajectory avoids static and dynamic obstacles such as forklifts, humans and pallets while keeping a fixed or minimum distance between the ATRs. Calculating the trajectory is done using NMPC using Proximal Averaged Newton-type method for Optimal Control (PANOC). The proposed behavior of the ATRs is adjustable by adjusting the weights in the NMPC formulations, and is able to change between safe and more aggressive behavior. Obstacles are modelled using polygons. In the NMPC formulation convex constraints can be handled, thus the polygon is split into multiple triangles using con-strained Delaunay triangulation resulting in a set of convex constraints. Formation control is done using a leader follower approach. First a trajectory is generated for the leader. It is then modified for the follower in such a way that the trajectories are at a fixed distance apart at each time step. The approach is evaluated through simulations and using physical ATRs by simulating a set of challenging scenarios. Most scenarios are handled very well by the proposed method, but for large dynamic obstacles the optimization problem might, in some situations, fail in finding a trajectory avoiding the obstacle. Also in scenarios when the leader has to adapt a lot to the follower it struggles. The physical evaluation support the evaluation done using simulations.