Distributed Estimation Under Local Information and Local Interaction
We are interested in distributed estimation under local information and local interaction. Our group designs and analyzes novel distributed algorithms to achieve fully distributed information fusion with multiple networked sensors in the face of noise under very mild assumptions on local observability, communication graphs, and models (time varying, nonlinear, and/or unknown). We address the problem of distributed joint localization and target tracking using a mobile robot network. Here a team of mobile robots equipped with onboard sensors simultaneously localize themselves and track a moving target in a distributed manner in a GPS denied environment. We also study the distributed average tracking problem and present distributed discontinuous algorithms to solve the problem. The central idea is that multiple agents estimate the average of multiple time-varying reference signals, each of which is available to only one agent, under communication with only local neighbors. The problem has applications in distributed measurement, feature-based map merging, and distributed Kalman filtering.
An experimental demonstration of our fully distributed joint localization and target tracking algorithm on the UTIAS multi-robot cooperative localization and mapping dataset can be found here. An experimental demonstration of multi-robot cooperative visual-inertial odometry can be found here (collaborative work with UD RPNG).
Distributed Optimization Under Local Information and Local Interaction
We are interested in distributed optimization under local information and local interaction. Our group studies the problem where multiple agents capable of communicating with only local neighbors cooperatively optimize a team objective function formed by a sum of local objective functions, each of which is known to only one agent. We have proposed distributed event-triggered mechanisms and addressed the case subject to non-identical constraints and communication delays. We have also studied how multiple agents with continuous-time physical dynamics are able to cooperatively achieve motion coordination and team optimization with only local information and local interaction. We deal with fully distributed design, finite-time convergence, time-varying cost functions, local constraints, and physical agent dynamics. Applications include distributed optimal motion coordination, multi-robot multi-target navigation, and distributed resource allocation.
An experimental demonstration of a distributed non-smooth optimization algorithm can be found here (robots forming a formation centered at the team optimal location). Experimental demonstrations of distributed time-varying optimization algorithms can be found here (the center of multiple quadrotors tracking a time-varying team optimal trajectory) and here (multiple quadrotors tracking a moving whiteboard).
Distributed Control Under Local Information and Local Interaction
Distributed Motion Coordination
One research venue is distributed motion coordination under local information and local interaction. Our group studies how a team of mobile autonomous agents can intercept or track a dynamic leader or target in the presence of reduced interaction and partial measurements and how a team of mobile autonomous agents can be used to maneuver or herd another team with only local interaction as well as experimental implementation and validation. We have addressed those problems under the realistic constraints that the dynamic leader (or leaders) is a neighbor (or neighbors) of only a subset of a group of followers, all followers have only local interaction, and only partial measurements of the states of the leader (or leaders) and the followers are available. Our group also studies how to guarantee that multiple autonomous agents achieve a geometric configuration or collective motions in a distributed manner and explores applications in planetary exploration, spacecraft formation flying, and cooperative patrol. We have developed novel distributed virtual-structure-based architectures for formation keeping and attitude coordination and explored collective period motions generated through Cartesian coordinate coupling and coupled harmonic oscillators.
An experimental demonstration of three leader robots herding two follower robots can be found here. An experimental demonstration of four robots following a circle while maintaining a square formation can be found here (static interaction topology with a directed spanning tree) and here (switching interaction topology with a directed spanning tree at each time instant). An experimental demonstration of multiple robots using coupled linear harmonic oscillators can be found here.
Another research venue is distributed consensus with applications in multi-vehicle cooperative control. Information consensus guarantees that agents sharing information over a network have a consistent view of the information critical to the coordination task. The problem is particularly challenging because of limited and unreliable communication/sensing. The basic idea for information consensus is that each vehicle updates its information based on the information of its local neighbors so that the final information of each vehicle converges to a common value. This basic idea can be extended to a variety of scenarios that incorporate group behavior and dynamics. Consensus algorithms have applications in rendezvous, formation control, flocking, attitude alignment, decentralized task assignment, and sensor networks. We have designed and analyzed distributed consensus algorithms for agents with various dynamics including first-order, second-order, general linear or fractional-order dynamics, rigid body attitude dynamics, Euler-Lagrange dynamics, and more general nonlinear dynamics. We have further studied various realistic issues in distributed consensus including sampled-data settings, optimality, time delays, adaptive design, limited information, state constraints, finite-time convergence, fully distributed design in asymmetric networks, and experimental implementation and validation.
Experimental results of the consensus algorithms on a mobile robot network can be found here (four robots rendezvous with a dynamic interaction topology that is directed switching but has a directed spanning tree jointly.), here (four robots rendezvous with a static interaction topology that has a directed spanning tree.), and here (four robots align their positions along a horizontal line with an even separation distance with a static interaction topology that has a directed spanning tree.). A simulation demonstration of multiple spacecraft attitude synchronization can be found here while an experimental demonstration on the ground robots can be found here.