Description

Introduction #

Cyber-Physical Systems (CPS), or Cyber-Physical Systems, are systems that integrate computation, networks and physical processes, with feedback loops where physical processes impact computational processes and vice versa. The main challenge in these systems is to combine abstractions that have evolved over centuries to model physical systems (e.g., differential equations and stochastic processes), with abstractions from computer science (algorithms and programs), which provide a “procedural epistemology”, i.e., they move from the “what is” notion of the experimental sciences to the “how is it done”.

From the experience of the MINA research group (Institute of Computing - Faculty of Engineering) in this area and in particular to the work in cooperative mobile robotics carried out in recent years, it has been identified as a need for the group to have a modeling and simulation environment for SCFs; in particular for cooperative robotic systems (systems of multiple robots with shared goals). The emergence of this need is multi-causal. On the one hand, having models and simulation platforms for SCFs allows for rigorous but simplified experiments that reduce costs and time. For example, it allows testing of new developments without the need for hardware deployment and reduces testing times. It also allows scalability testing by increasing the size of the scenarios or the number of agents, aspects that would be unfeasible (due to space or cost) in a physical test environment.

Finally, a modeling and simulation environment is an indispensable part of the digital twins concept. Digital twins offer a novel way to test the impact of different control decisions in a simulated virtual environment without requiring changes to the physical system. In this way, various control options can be calculated and tested in parallel to the execution of the physical system.

Being able to have a simulation platform that can correctly model all aspects of an SCF is complex and presents several challenges. The capabilities and skills needed to correctly describe an SCF are very varied and come from different application fields. In the case of simulation, each of these fields in general have their own established and reliable models and tools. The major challenge is then to be able to harmonize all these heterogeneous points of view and to integrate the tools and models of each domain into a single simulation environment.

It is in this context that the concept of co-simulation appears as a strategy to solve these challenges. Co-simulation consists of reusing models and tools implemented in different simulation platforms and making them interact with the objective of obtaining a new simulation platform that contains them. In other words, co-simulation allows the global simulation of a system that integrates physical, software and communication network aspects through the composition of different simulators specific to each of these areas. In this sense, each individual simulator is seen as a black box capable of implementing a certain behavior, consuming inputs and producing outputs. To achieve this composition it is necessary to manage the exchange of information between the different platforms as well as to synchronize their execution.

This project aims to study the SCF co-simulation problem in general, but with a specific focus on the co-simulation problem and the digital twins for cooperative mobile robot systems with wireless communication. For this, it will be necessary to deepen both in the theoretical aspects of hybrid system simulation (modeling physical and computational systems) and in the practical aspects of implementation of these co-simulation platforms.

In addition to the expected theoretical results, they will be translated into a co-simulation platform that will serve as a common basis for the research in cooperative mobile robotics of the applicant group. This will open a new line of research in the MINA group that will allow it to advance its research plan in SCF, while training new human resources in this area.

Description of the research problem #

The research problem of this project arises from the need for a simulation platform for mobile cooperative robots. Robots, for the purposes of this project, can move in, and interact with, the physical world without human supervision. To benefit from the use of a multi-robot system, coordination strategies between agents must be designed to effectively implement cooperative solutions. In this regard, network communication is a key factor. For a team of mobile robots, the wireless network through which they communicate is not only a constraint of the problem to be solved, but another object that the team can manipulate and use as a tool, modifying it according to their needs, altering various transmission parameters as it is usually done but, above all, modifying the network topology as the robots move. Having a simulation environment, such as the one in the Figure, that can accurately model both robot movements and control as well as communication behavior is very important in the research and development process of this discipline. Moreover, in the context of digital twins, this accuracy is critical since simulations are used to test, validate or learn aspects of the system that will later be used in the physical environment. In addition to the accuracy and reliability of models and simulations, in the context of digital twins, time constraints are added. That is, simulation tools must allow processes to run faster than in a real environment. To address this problem in a feasible and scalable way, co-simulation presents itself as a possible solution albeit with a significant list of challenges.

The research problem of this project, is ultimately to solve the list of challenges presented by hybrid co-simulation in general in the particular case of cooperative robot simulation with runtime constraints that allow for statistically meaningful studies and to integrate the digital twin concept to control loops. The generic challenges of hybrid co-simulation are summarized in \cite{Gomes2018Co-Simulation}, below we list those that will be faced in the context of this project.

**Semantic adaptation and model composition. Each simulator involved in a co-simulation defines its own data, interface and semantic models. For the purpose of making them interoperate, although generic wrappers can be used to provide them with common interfaces and models, in fact, the abstraction needed to achieve this depends on the specific scenario. There is simply no single adaptation option that is best in all cases. Even at the technical level, the way in which events or signals are sent to (or obtained from) each simulator may need to be adapted.

**Predictive step size and event location **In the hybrid approaches we consider in this project, the time advance has to be explicitly defined. However, an inadequate time step size may cause loss of events from one or the other simulator. There are several approaches to this problem in the literature, but, up to this point, we do not know of a suitable solution for our particular scenario and objectives in which continuous processes at the radio-frequency or physical motion level in a continuous-time simulators must interact with event-based messaging and decision making processes in discrete-event simulators. As an additional constraint to the usual ones in hybrid co-simulation, in the case of our project we add the requirement to choose steps of a size that allow simulations to be performed at a speed compatible with the digital twin concept.

**In a co-simulation, a continuous-time simulator can generate, through its wrapper (wraper), a sequence of output events corresponding to the sampling of a given signal. However, from the point of view of another of the simulators involved, it is not possible to discern whether it is a sampling of a continuous signal, or discontinuities of a sequence of discrete events. This type of ambiguity must be resolved by adding functionalities and extending the information models of each simulator. Once identified, these discontinuities must be handled correctly. Continuous-time simulators are, in reality, a mock-up of a continuous system and discontinuities in the inputs must be handled so that the resulting simulation makes sense in the physical world it simulates, e.g., total energy must be conserved, and a robot must not cross a solid surface.

**In the case of a hybrid co-simulation, it is possible for a particular sequence of events to cause the system to become unstable, even if all the continuous elements of the simulation are stable. It becomes necessary to identify, in the case of cooperative robots, whether the continuous-time simulators involved may induce unstable trajectories as a result of noisy inputs, formatting, delays or messaging frequency.