Our research vision is to build robot systems that are capable of acting competently in the real world. We study, develop and combine automated planning and control methods to achieve this capability. We strive to implement these methods as publicly available software solutions. These serve as tools for deepening our theoretical and practical understanding of planning and control for single- and multi-robot systems; and as frameworks for developing automation solutions in cooperation with our industrial partners.
Both discrete and continuous models are relevant in systems composed of robots, e.g., continuous time and discrete events, metric maps and qualitative spatial relations, kinodynamic motion models and symbolic preconditions for acting. Methods for automated reasoning differ significantly depending on whether the given models are discrete or continuous. How should discrete and continuous models be synchronized, and how can they be reasoned upon jointly?
Planning and control methods offer means to solve problems in a general way, but require models that are abstractions of the real world, e.g., discretizations of the environment in motion planning, linear robot dynamics in control, contiguous time intervals in scheduling. Which are the most suitable forms of abstraction for planning in different multi-robot applications? Can abstractions be inferred automatically from continuous data?
Synthesizing competent behavior of multi-robot systems in the real world requires reasoning about requirements pertaining to heterogeneous aspects of the multi-robot system and its operating environment, e.g., temporal and spatial constraints, kinodynamic constraints, models of resource consumption, and criteria for allocating tasks to robots. How can we develop algorithms for searching in the space of several of these requirements jointly?
Multi-robot systems epitomize the need for planning methods, which provide them with the means to cooperate, achieve goals, and move in a coordinated fashion in the environment. Yet the physical nature of multi-robot systems, as well as the uncertainty connected to robot behaviors and perception, destroy many of the assumptions made by current planning methods. How can we ensure the correctness, completeness, and efficiency of multi-robot planning methods under realistic conditions?
In many real-world applications, decisions must be made online, while previous plans are already under execution. These instances of (re-)planning may be local, concerning one or few robots, or global, pertaining to the entire multi-robot system. How can we guarantee the consistency of decision-making in such settings? How can we automatically identify the extent to which we should replan for a multi-robot system? How can we guarantee the correctness, safety and optimality of online planning?
A multi-robot system may be regulated by a centralized decision-making agent, or decision-making may be distributed among the robots in the system. Centralized approaches can leverage a global view to better optimize, while distributed approaches can exploit locality to keep the computational burden low. When is it convenient and necessary to adopt a centralized or a distributed modality? How to build multi-robot systems that can transition between the two as appropriate?
Multi-robot systems often include robots with different skills. These skills may change over time, as new features are added to existing robots or robots are added and removed from the fleet. How do we build high-level fleet planning methods that do not depend on the specific planning and control schemes needed for the individual robots in the fleet? How can we build a general, robot-agnostic, high-level control loop that can be instantiated in different concrete multi-robot systems?