Back to main page of HARMONY

Project tasks


Task 1 - Project management and dissemination

Task coordinator: INESC-ID

This task aims at monitoring and controlling project execution, thereby ensuring that the work flows according to what has been planned. This activity includes the dissemination of project results. In its material and scientific components, project management and coordination is ensured by the IR, by the co-IR, and by a commission that includes one representative from each of the partner institutions. This commission ensures the coordination of task 7, that relates to the articulation of project activities with the intelligent strategies at the national and regional levels for Norte and Lisboa.

Task 2 - Dynamic programming based optimal control problems

Task coordinator: FEUP/SYSTEC

2.1 - This task focus on the study of complex engineering problems using modern methods of applied mathematics and numerical schemes for dynamic optimization. In order to use powerful and efficient computational methods to study distributed optimal problems of interest we will pursue:

- alternative reformulations that can be successfully studied via mathematical and computational methods.

- the use of tools from optimal control theory will be used to gain extra insight on the problem. In this respect a novel aspect will be reachability analysis; it allows the study of feasibility of a mission and allows to isolate hazardous areas to avoid.

- the analytical study of optimal syntheses of the problems or sub problems of interest. This aspect is of foremost importance to get to characterize the profile of the solutions.

- definition of value functions for the problems studied.

Computational tools to be used in this task include: Nonlinear Programming Solvers like Ipopt and WORLP interfaced with AMPL. Note that WORHP and IPOPT are both free for academic purpose and SYSTEC holds a license for AMPL. Optimal Control Solvers: BOCOP and ICLOCS HJB based software: ROC-HJ and other known solvers. The Hamilton-Jacobi (HJ) approach provides a powerful method to characterize the reachable sets as the level sets of the value function of optimal control problem. In some cases, HJ solvers permit the reconstruction of trajectories (as it is the case with ROC-JH)

2.2 - Hamilton-Jacobi-Bellman (HJB) type of conditions for impulsive control systems. This sub-tasks concerns mathematical control systems which can be regarded as an abstraction model of control systems exhibiting very fast and very slow dynamics. The motivation arises from many contexts in which the control and optimization of the fast and of the slow dynamics cannot be decoupled such as in joint operations of aerial, surface and underwater vehicles for precise data sampling and timely rendezvous operations. The measure-driven dynamics framework will be adopted and challenges to be investigated include:

(i) appropriate solution concept, conditions under which solution exists, conditions under which the value function is solution to the HJB equation,

(ii) relations between the adjoint equation of the Maximum Principle for impulsive control systems and the solution the associated HJB equations, and appropriate formulation of impulsive feedback control

(iii) Hamilton-Jacobi-Bellman (HJB) type of conditions for systems for sampled impulsive control systems.

The approach envisaged to address the above challenges consists in the appropriate approximation of the impulsive control system by a sequence of conventional ones and determine both solution concept and set of assumption under which a meaningful sense can be obtained for the HJB equation for impulsive control systems. Methods of Real and Functional Analysis, Optimal Control Theory, and Measure Theory will be adopted.

Task 3 - Algorithms for distributed optimal control

Task coordinator: INESC-ID

The methodology to follow consists of decomposing the overall control problem in a set of local interacting dynamical models, together with a set of local objective cost functionals that add up to the global functional. The solution of the local optimal control problems is to be done in such a way as to approximate tightly the solution of the global optimal control problem. We propose to solve this distributed optimal control problem by combining two main steps. In the first step, the solution of the necessary conditions of optimality given by the maximum principle, or equivalent are approximated using a convenient numerical method for optimal control. In the second step, the resulting finite dimensional optimization problem is solved using the methods of distributed optimization. For both the above steps different possibilities are envisaged, depending on the detailed formulation. For the first step one may use either different versions of the maximum principle that depend on the specific problem (such existence or not of state constraints, or the space in which the control functions are assumed to exist), or methods from Approximate Dynamic Programming to allow feedback solutions. For the second step, a variety of methods from distributed optimization are envisaged, such as distributed ADMM (alternating directions method of multipliers) or methods from game theory. A key aspect is therefore distributed optimization. In many practical applications, and opposite to the current literature that considers network-oblivious type algorithms, the network topology is well known in advance. For such cases, we should do better, and for that sake this task proposes to design distributed algorithms that are network-aware. That is, we tailor the algorithm at design time to a given network, boosting performance since the algorithms speed increases. To tailor the algorithms to target networks, we propose to exploit a usually neglected parameter in distributed algorithms ? the weight matrix. This matrix weights the messages between neighbour agents, controlling the flow of information. The matrix is typically pre-chosen without considering two important pieces of information: the network topology and the class of functions to be minimised. We suggest marrying the weight matrix with this available information, and, more importantly, make the weight matrices time-varying. Making the matrices time-varying not only increases the number of parameters to be found (one matrix per time instant) but also makes the problem of finding the optimal weight matrices nonconvex (those matrices multiply each other across iterations, and such multilinear products induce a nonconvex structure). These fascinating challenges require new optimisation approaches for which our team is well-prepared.

Task 4 - Distributed MPC for networked cyber-physical systems

Task coordinator: FEUP/SYSTEC

This task is dedicated to R&D of theoretical tools and algorithms in the framework of distributed Model Predictive Control (MPC) for networked Cyber-Physical Systems (CPS). The control of networked CPS with performance and robustness guarantees poses considerable theoretical challenges and opportunities since the interactions between the different components of CPS are closed through limited communication channels.

4.1 Distributed estimation and localization of networked CPS with uncertain relative measurements and communications: Most of the existing studies assume implicitly that the localization problem is perfectly solved. The goal here is to develop a joint localization and estimation framework with formal guarantees on the estimation error, obtaining iterative optimization-based estimation algorithms that improve the overall target estimate as measurements are acquired and communicated.

4.2 Distributed MPC for coordinated output regulation in multi-agent systems. Given a general multi-agent system where each agent is characterized with a continuous nonlinear dynamical model, an output equation that is a function of the state of the agent, and a coordination vector that depends on the coordination states of the neighboring agents, the goal of this task is to study and develop design distributed control strategies that steers the output signals to a desired steady state, while simultaneously driving the coordination vectors of the agents to consensus by minimizing a performance index that combines the output regulation objective with the consensus objective.

4.3 MPC Cloud-based Control of networked CPS. The aim is to combine the potentialities of cloud technology and MPC to create a framework for the coordination and control of networked cyber-physical systems. Key points to addressed include the study and design of advanced event-triggered control strategies combined with MPC tools; the analysis of stability, convergence and robustness of the proposed control algorithms in the presence of limited communication and computation resources; and the development of sw modules with the implementation and performance illustration of the proposed algorithms.

4.4 This tasks concerns the coordinated control of multiple agents in networked CPS for which the control and estimation processes do not depend only on the data that is originated in the the agents, but may also depend on data originated from an external context. This requires an extra supervisory control layer, motivated by the fact that the operation of a networked system in which each node tries to achieve its own best performance, results in a collective performance that might be far from the best achievable global performance. To show convergence, stability, and robustness results, we consider methods of Invariance, Lyapunov, OC Theory, and Functional and Real Analysis.

Task 5 - Energy management with multiple renewable resources

Task coordinator: INESC-ID

In this task a case study is developed in relation to energy production and storage systems in which the distributed optimal control techniques developed in the project are applied. One considers a system in which there are multi-generators, including solar thermal energy for electricity generation, storage capacity in the form of a heated fluid, and multiple consumers. Since it is assumed that each generator cannot meet the consumption targets alone, while each subsystem is provided with a local control system, they must cooperate in order to achieve the overall objectives. The absence of a centralized manager refers to a distributed control problem, which motivates the use of the methods under study. Regarding the performance of the global system, several scenarios are compared regarding the topology of the graph that reflects the possibility of communication between the various subsystems, and the possibility of limitations in varying degrees of the amount and rate of information transmitted. The main challenge consists in developing a coordination protocol that involves the minimum amount of information being interchanged among the local control nodes that manage the energy network, and such that the global performance approximates as much as possible the performance achieved under centralized optimal control and perfect communication.

Task 6 - Multiple vehicle control and coordination

Task coordinator: IST-ID/ISR

Impressive flight formation capabilities, including transitions between different formation patterns, have been experimentally demonstrated in indoor environments, taking full advantage of external motion capture systems, which provide full state measurements at high-frequency rates [TUR14, AUG12]. In these examples, the number of vehicles is typically fixed and flight time restrictions are not taken into account. Flight endurance together with payload can be considered as the main limitations of multirotor vehicles. To achieve increased autonomy, we aim to consider formation control problems where vehicle replacement needs to take place. The goal will be to pose the problem as an optimal control problem so that vehicle replacement has minimal impact on the formation and does not disrupt the mission being performed. This might involve different objectives such as ensuring connectivity and collision avoidance during the transition, adequate scheduling of replacements based on battery monitoring, or maximizing area coverage of the ensemble when performing survey missions. A different objective will be to define optimal control problems considering cost functions that are not only local to each vehicle but are also defined in local reference frames. This approach draws inspiration from distance-based formation control [OPA15], where individual vehicles can only sense the relative positions of their neighbors with respect to their own local reference frames, which are not necessarily aligned. It is particularly suited for application in GPS-denied environments, where the vehicles must rely on local sensors such as vision or laser to acquire information about the surrounding environment, including the relative position of their neighbors.

Task 7 - Integratyion with national and regional strategies

Task coordinator: IST-ID/ISR

This task aims to guarantee and enhance the valorization of the project results from the point of view of the National and Regional Intelligent Specialization Strategies of Lisbon and the North. To this end, a set of actions will be developed to link the research team with business stakeholders who have the potential to endogenize and develop the results of the project, transforming them into innovative products, or incorporating them into their own products in a way To achieve added value and competitive advantages. These include regular contact with professionals from selected national companies in the areas of Information and Communication Technologies, including software producers for automation control and robotics, companies in the aerospace sector active in the development of avionics systems, and navigation and control systems for the Space, and companies in the energy sector. In addition, a seminar will be held dedicated to companies, and a text explaining the main results of the project in a language accessible to engineers who work in companies, with emphasis on the results with the greatest application potential. From the point of view of the strategic areas, at the national level, the emphasis will be given in Axis 1 - Information and Communication Technologies, particularly in the topics Internet Promotion of Future, Advanced and Complex Engineering Systems; Energy, in the topics Optimization of the Production and Transportation of Energy and Application of New Technologies and Intelligent Energy Networks, highlighting the applications related to energy storage. And still in Axis 3 - Automotive, Aeronautics and Space in the topic Development of Advanced Technologies Applied to the Automobile, Aeronautics and Space. The activity in these topics is articulated, and naturally integrated, in the pursuit of scientific objectives of the project included in the intelligent specialization strategies of the North (field of specialization: Resources of the sea and Economy, with regard to the articulation between submarine robotics with Coordinated mobile robots and monitoring of marine resources) and Lisbon (with regard to the development of innovative controls in the aerospace and energy fields, with a potential impact on mobility and transport for multiple autonomous vehicles).

Task 8 - Final report

Task coordinator: INESC-ID

The aim of this task consists in the preparation of the final scientific report presenting an overall synthesis of the main achievements of the project. It will be coordinated by representatives from all the partners. The main result is the comprehensive and critical description of the distributed controller structures developed and its main design guidelines proposed for cyber-physical systems.