Experts in robotics, communications systems, multi-agent systems, sensor networks, applied AI, embedded systems, and autonomous systems met in the MINA research group. In short, the group brings together researchers from the main Cyber-Physical Systems disciplines (CPS), which integrate computing, networks, and physical processes, with feedback loops where physical processes impact on computation processes and vice versa.
The main challenge is to combine abstractions that have evolved over centuries to model physical systems (e.g., differential equations and stochastic processes), with computer science abstractions with some decades of evolution (algorithms and programs), which provide an "epistemology of procedures", going from the experimental sciences notion of "what is it" to that of "how it is done".
Without pursuing it, the MINA group has faced most of the CPS methodological difficulties since its creation in 2003, based on the initial experience in mobile robotics and network management (hence the non-acronym MINA - Network Management / Artificial Intelligence). Based on this experience, the development of a research line specifically dedicated to the fundamental aspects of the CPS has been identified as a necessity, as well as another devoted to IoT (a CPS emerging application), and the development of its existing lines under the methodological framework of the CPS. The economic and social potential of these systems is enormous and is attracting attention from industry and academia worldwide. Our research work seeks to have an impact on areas in active development and the social and productive impact such as automation, IoT, smart cities, agriculture, among other fields of application, which are beginning to materialize in our country.
The growing traffic demands that future wireless networks will have to deliver, in many cases several orders of magnitude greater than presently, present enormous challenges. In this context, appropriate wireless network management is essential not only to meet connectivity requirements but to make efficient use of resources. The management of wireless networks includes various aspects such as transmission parameters, and the allocation and management of network resources.
The management of transmission parameters aims to maximize the transfer rate while minimizing the interference and saving energy. To accomplish such objectives is necessary to study the relationship between the different transmission variables and how their correct management can improve performance. Examples of these variables are the transmission power, the transmission rate, and the sensitivity of the media access mechanism. Due to the complexity and dynamism of the transmission media, where the optimal configuration changes continuously as the actors of the system move and the radio conditions of the environment change, this management is both essential and complicated. Therefore, research in this line requires the theoretical study of network behavior, the use of mathematical models as well as an approach based on machine learning techniques.
Regarding resource management, the main objective of this research sub-line is the design and development of the techniques and mechanisms necessary to achieve an efficient allocation of resources in the context of wireless and heterogeneous access networks. Usually, these tasks are part of the planning of the access network, but the dynamism and intensity of use of wireless networks have driven the idea of automating management by bringing it closer to the field of control theory under the name of self- management or autonomous management. The group's work in this line has focused especially on autonomous and distributed control and management solutions.
Navigation is one of the most important and challenging activities that must be taken into account when working with a mobile robot. Navigation puts togheter very aspect of robotics: sensing, acting, control architecture, planning, problem solving, computational efficiency and hardware. Navigation is a collection of algorithms that allow solving the difficulties that appear when the robot try to answer the following questions:
Where should i go? This problem is usually determined by a human or mission planner of the robot control architecture.
What is the best way to get there? This problem is called trajectory planning and is the navigation aspect that has received most attention.
Where I have been? It refers to the construction of maps, one of the aspects of navigation that has been overlooked in the past but has gain momentum in recent years.
Where I am? The robot must know where it is to follow a path or build a map. This aspect of navigation is called localization.
Some of these problems are solved simultaneously in pairs, for example, exploration, active localization and SLAM (simultaneous localization and mapping). The objective of the exploration is to cover the entire environment in the shortest possible time. Exploration assumes that the robot can locate itself, having to build the map and plan trajectories to achieve its objective. In the case of active localization, it is possible to make movement decisions to improve the efficiency or robustness of the localization. Map construction and localization are strongly related and inter-dependent, which has led to their joint study under the name of SLAM. Map construction introduces the need to locate while the robot moves in the environment. SLAM refers to the problem in which a mobile robot builds a map of its environment while simultaneously locate itself within that map.
As a result of the evolution of both, image-capture devices and processing algorithms, there is a great interest in conducting guided navigation for object identification. The most widespread methods for dealing with navigation are, in their nature, probabilistic. There are also some alternatives based on biological models, e.g., based on rodent studies. Some of the works developed in this line of research are: robust navigation based on the behavior of rats, outdoor navigation for support in apple harvesting tasks supported by artificial vision and exploration of environments with unmanned land and air vehicles.
This line of research refer to infrastructure topics in general, comprising two main subjects:
Management and Control of resources in virtualized networks.
Design and Optimization of Intelligent Networks.
Management and Control of resources in virtualized networks.
Next-generation networks involve ubiquitous services, heterogeneous technologies and moving users in an intelligent environment (for example, the so-called “smart cities”). That presents a challenge for both access networks (with the prevalence of wireless technologies, and residential broadband access), and for the global Internet, which serves as an application transport infrastructure for users who demand quality of experience. These scenarios differ fundamentally from the founding premises of the Internet. In this sense, global content providers seeking to improve the user experience have deployed infrastructure (datacenters) with their content replicated in numerous locations forming "Content Distribution Networks" (CDN). CDNs have their resource management challenges, but also impose challenges to the global routing system, and, in particular, to BGP (Border Gateway Protocol), the protocol that keeps the network communicated and allows the information to reach the end-user.
The need to deploy services dynamically and with low operating costs has accentuated the prevalence of software, and, in particular, the virtualization of execution platforms and/or network resources, leading to popularize the concept of "Cloud Computing": applications run in a virtualized environment and data is stored in datacenters replicated in various geographic areas, as mentioned before, requiring dynamic cloud resource management. Network devices are intended to be generic and programmable, giving rise to the concept of "Software Defined Networking" (SDN). On the other hand, telecommunications operators seek to be able to orchestrate network services by combining basic functionalities dynamically; the concept is called "Network Function Virtualization" (NFV).
Design and Optimization of Intelligent Networks.
Network design and optimization is a discipline with many fields of application, i.e., telecommunications, the interconnection of microprocessors, transport problems, and electrical network, among others. This line of research is strongly based on the problems of telecommunications networks, particularly in flexible optical networks (flexigrid), whose fundamental characteristic is the elastic and dynamic allocation of spectral bandwidth for each link, without need to use fixed wavelengths. Current networks form a decentralized multi-domain structure, which makes it difficult to coordinate optimization and reconfiguration tasks. These tasks are mostly performed manually by experts, and there is a lack of programmability and automatic feedback loops to perform these tasks. A fundamental cause of the inefficiency of current networks is the absence of "cognitive intelligence," that is, the ability to infer the state of the network, analyze the implications, and take actions proactively. Intelligent networks should have "Self*" capabilities: Self-Aware, Self-Configuration, Self-Healing, Self-Optimization, which know the information at different levels (Infrastructure, Services, Flows), and provide abstract intent-based services interfaces. Monitoring is a central activity to ensure these characteristics, within “Observe-Analyze-Act” cycles, seeking to reach automated and scalable control and management plans that have both local (distributed) and global (centralized) knowledge. Starting from the flexigrid optical networks use-case, the main objective of this research line is to study the evolution of the networks towards an intelligent and "self-aware" infrastructure. It also seeks to explore other use-cases, along with training of human resources in the area and cooperation with local research groups and related actors (e.g., public and private companies).
We have been working on conceptual aspects of this line of research for some time, for example, in the problems of wireless network management, and in the development of embedded systems, which are basic components of an IoT solution. We have also been working in sensor networks, machine-to-machine communications (M2M), and device-to-device communications (D2D), all predecessor concepts of IoT. However, once the IoT concept has been widely established, many research opportunities emerge. In this sense, both in the industry and in the academy, a de facto architecture has been established, consisting of final devices (and gateways) that generate information in their sensor role and receive commands in their actuator role. This information is transmitted to/from public or private cloud IoT platforms through a specific asynchronous communication component (called broker), which makes data available to storage modules and analysis, which eventually take action. In the last couple of years, we have been evaluating the state of the art of these architectures, including platforms, application development patterns, among other challenges. Likewise, we have begun to study IoT security aspects, in cooperation with the "Computer Security Group" at the Computer Science Institute (GSI-INCO), focused on a Smart Grids security (a project with the state-owned electrical company), which includes smart meters on an IoT solution. Participation in this project has helped to grab the state of the art, and explore the security and safety aspects that appear in this type of applications, generally invasive of public and private space (for example, health care applications).
This line of resarch seek for architcture alternatives for the "cloud IoT" model, also looking for concrete solutions to specific aspects that contribute to improve the performance of IoT solutions. We also seek to develop theoretical aspects at different levels of architecture, which undoubtedly relates to oher lines of reasearch of the group, in particular with Decision Processes. Indeed, where (in which components) and how (by what algorithms or processes) decisions are made in an IoT solution, constitutes a central reasearch question.
CPS include devices that typically measure (sense) environment variables. For example, signal level and signal-to-noise ratio in the case of a wireless Access Point, temperature, or other environmental variables when speaking about exterior sensors, or distance to obstacles in the case of a robot. These measures, combined with other computational elements and information, and with high-level rules or objectives, feed various decision processes, for example modifying the transmission channel of an Access Point, balancing certain part of the traffic in the network, activate the irrigation in a plantation, or change the direction of the robot.
These decision processes, regardless of the application, constitute a specific area of research. In addition to our work on classic decision processes in robotics-related projects, our group conceived the RAN autonomous decision model for wireless networks, and subsequently adapted it to opportunistic networks (RON). There are also frameworks such as ROS (Robot Operating System) and OROCOS (Open Robot Control Software) that favor the development of distributed architectures for the control of autonomous agents, and the "cloud IoT" decision model.
We are interested in studying various aspects of the decision processes, including the classic centralization vs. distributed dichotomy, rule-based mechanisms or, in general, policies, fuzzy logic, machine learning, neuro-diffuse processes, consensus and election in distributed systems, cooperation and coordination, among others.
Some examples include research on rule conflict resolution, studies of stability of decisions based on fuzzy logic and other purely computational aspects of decision processes. We are currently working on the incorporation of machine learning into decision processes in radio interference control processes and in aspects related to network and computing infrastructure.
Distributed mechanisms and task selection algorithms have been implemented using OROCOS, to support the collaborative decision-making in fleets of robots that explore previously unknown environments. These systems can be used in a wide range of applications such as surveillance, patrolling, cleaning, agriculture, exploration of hazardous or inaccessible areas.
Moreover, ROS has been used to solve decision-making in various robotics scenarios, particularly the development of DM3 (autonomous vehicle to support fruit collection tasks) and SuperM (vehicle to assist wheelchair travel within supermarkets).
Service robots are increasingly present in our lives. In this sense, they must safely interact with people, anticipating the effect of their actions considering other agents that surround them, paying special attention to the actions and needs of the people with whom they share the environment. These agents must have natural language processing and speech generation capabilities, a complex mechanism for decision-making and learning, in a dynamic, stochastic, and real-time environment. Robots with these abilities have been called cognitive robots, bringing together scientists from different fields linked to cognitive sciences (computer science, neurosciences, psychology, and artificial intelligence), generally using models based on the cognition of living beings. Cognitive robots achieve their objectives by perceiving their surroundings, paying attention to the relevant events, planning what to do, and learning from the interaction with the environment. They deal with the inherent uncertainty of real environments through learning, reasoning and exchanging their knowledge with other agents.
In short, it is a classic area within several engineering disciplines but, since it is a cross-disciplinary aspect of CPS, it acquires a new interest in light of the new technological possibilities, considering the potential growth of CPS in the years to come.
CPS foundations include studying design tools that adapt to the physical and the cyber world and allow the development of design methodologies that work at the same time for both worlds. It is necessary to model computational systems that work at the same time on continuous and discrete signals and networks that communicate synchronous and asynchronous systems, considering the temporal variable as a critical element. Moreover, managing the scalability, complexity, and heterogeneity of CPS is a tough challenge. It calls for classic engineering strategies, such as modularity and reuse of elements of a system, and the ability to interoperate with legacy systems. To complete the complexity of these elements, CPS usually have very demanding correctness requirements and need complex testing techniques, simulation, and stochastic modeling, together with high safety and protection requirements that can only be ensured using formal methods techniques.
Our group cannot cover all aspects of CPS foundations, and therefore collaboration with other research groups is criucial to pursue this line of research in aspects related to modeling, scale management, simulation, continuous and discrete systems and safety modeling.
Instituto de Computación
Facultad de Ingeniería, UdelaR
Julio Herrera y Reissig 565