Project description
Multi-agent control using a leader-follower hybrid approach
A current trend in multi-agent control is the use of formal verification to identify more general task specifications that induce a sequence of control actions. However, in existing solutions, the quantitative transient constraints or high computational costs of controlling the whole agent group when the number of agents grows aren’t considered. The EU-funded LEAFHOUND project proposes a heterogeneous, leader-follower approach as an alternative. This relies on three design stages: deriving transient controllers for cooperative control objectives in leader-follower networks, employing controllers for tasks as spatio-temporal logic specifications, and tackling task dependencies and infeasibilities by using notions of network controllability. This approach opens up new avenues in the fields of control, robotics and formal methods-based synthesis.
Objective
Multi-agent control is a popular research topic due to its applications in a variety of areas. The main approach has been to incorporate tools from single-agent control to the multi-agent setup. However, many applications involve more complex tasks that may not be cast as a classic control objective, while the agents may be subject to constraints in space and time. A current trend is thus to use formal verification in order to specify more general task specifications that induce a sequence of control actions rather than a stand-alone objective. Spatiotemporal logics are based on continuous time signals and allow formulating space and time constraints, and thus are suitable for defining such specifications in multi-agent systems. Existing solutions do not take into account quantitative transient constraints, nor consider the high computational cost that may be endured by controlling the whole agent group when the number of agents grows. We instead consider here a heterogeneous, leader-follower approach, which relies on three design stages. At a first stage, we derive transient controllers for cooperative control objectives in leader-follower networks; at a second stage, these controllers are employed in order to satisfy tasks given to the network as spatiotemporal logic specifications; and at a third stage, we tackle task dependencies and infeasibilities through a refinement process both in the transient control design as well as in the leader-follower group structure, by using notions of network controllability. The proposed leader-follower approach to the spatiotemporal task planning problem, combining elements from cooperative multi-agent systems under transient constraints, formal verification and graph theoretic network controllability, requires for new ways of thinking and approaches to analysis and design; it thus constitutes the proposal a beyond the SoA and groundbreaking approach to the fields of control, robotics and formal methods based synthesis.
Fields of science
Keywords
Programme(s)
Funding Scheme
ERC-COG - Consolidator GrantHost institution
100 44 Stockholm
Sweden