Coordination and planning (joint Örebro/Pisa project)
Pioneered by web giants like Amazon and accelerated by ever-increasing e-commerce demand, automated logistics solutions have been progressing at high speed in the last ten years. However, a standing challenge is to effectively yet safely coordinate large-scale heterogeneous multi-robot fleets in real applications while accounting for uncertainties and contingencies. Most existing approaches require constraints on the infrastructure or unrealistic assumptions on robot models. The following projects aim to overcome these limitations. The proposed work will involve the use of the coordination_oru library, a general-purpose tool for integrated motion planning, coordination, and control which is used in several research and development efforts in cooperation with industrial partners (including Scania, Epiroc and Volvo). More information about the coordination_oru library can be found at https://github.com/FedericoPecora/coordination_oru.
In the following, we describe three project proposals aligned with the above research goals. A suitable MSc thesis may comprise of e.g. projects 1+2 or 2+3. Please get in touch with the contact persons listed below for more information.
All the projects proposed in the following may require an international collaboration with other students attending Örebro and Pisa Universities.
Project 1: Limited horizon lookahead on precedences for liveness
The current approach provides one global and one local strategy for deadlock prevention. The former is complete but has exponential complexity in the worst case. The latter is incomplete but has polynomial complexity. In this project, we seek to achieve a better trade-off between complexity and completeness, by using prior knowledge of robot kinodynamics.
Project 2: Limited horizon path knowledge
The current approach is based on decoupled motion planners (i.e., each robot computes its own path to the goal). However, there are situations in which only “a few steps of the future path” can be known in advance (e.g., the goal may not be available for privacy reasons or the path may be computed using receding horizon motion planners). In this project, we wish to investigate how such situations can be included in the framework while preserving both safety and liveness.
Project 3: Distributed precedence based coordination
The current approach relies on a centralized fleet coordinator and the ability to communicate between it and all agents. In some cases, however, it is unnecessary to control the entire fleet from a centralized coordinator; in others, this may be impossible due to privacy and/or communication limitations. In this project, we seek to investigate distributed alternatives to centralized coordination. An important research question here is to identify the minimal information that needs to be shared between agents in order to guarantee safety and liveness via precedence constraints in a distributed fashion.
- Excellent programing skills in Java and/or C++.
- Prior knowledge and/or strong interest in motion planning, optimization, control, and multi-agent systems.