Fall 2024 GRASP Seminar: Knut Åkesson, Chalmers University of Technology, “Factory with Vision – Planning and Control of a Bird’s-Eye-View-Guided Mobile Robot Fleet”

At Chalmers, we are collaborating with a major industrial partner to develop a control system for a fleet of mobile robots, guided by a network of cameras providing an overhead view of the operating area. In this talk, I will discuss the challenges involved in implementing such a control system. These challenges include the perception system, which requires fusing data from multiple cameras into a coherent representation of the current state. Additionally, the perception system is used to predict the system’s near-future state, which aids in trajectory planning for the fleet.

Another key challenge lies in high-level planning, which involves scheduling the fleet to ensure that robots reach specified locations within defined time windows, selecting optimal routes, and managing battery levels. The planning phase accounts for static obstacles but not dynamic ones, as it generally occurs before information about dynamic obstacles is available. Instead, dynamic obstacles are handled during the trajectory generation phase.

In the talk, I will also present recent advances that combine deep reinforcement learning with distributed model predictive control to enable effective trajectory generation for a fleet of mobile robots, even in the presence of dynamic and potentially non-convex obstacles.