RICERCA ITALIANA     

Control Management of Cooperative Heterogeneous Unmanned Vehicles: Analytical Methods and Experimental Validation

Università di Pisa

Abstract

The objective of the proposed research is to develop a methodology for the management and control of heterogeneous vehicles (land
and air for instance), so that they are capable of operating autonomously and in a cooperative fashion. In addition, the methodology
will be tested experimentally by selecting a scenario consisting of up to three vehicles (two air vehicles or UAVs, and one ground
vehicle or GAV). The problem of managing the mission of several autonomous vehicles, possibly of different nature, such as land,
air, and sea vehicles, has become of enormous interest from the standpoint of basic as well as applied research, for a variety of
potential applications, included its potential contribution to defense and prevention against terrorist attacks. In this field, platform
technology has dramatically improved in recent years (material properties, miniaturization of components, energy storage, computer
speed and storage capacity,..), however reliable and flexible high level automation and "intelligence" are still open questions. The
aim of the project is to contribute to the latter aspects in terms of trajectory planning optimization in the presence of static (known)
and partially unknown obstacles, to the definition of appropriate sensor suite and sensor fusion, and to the evaluation of
communication links and their fault tolerance requirements. In order to test the developed methodologies, an experimental scenario
will be set up representing a typical search and rescue operation. <<<

Principal Investigator

Mario INNOCENTI Università di PISA

Research Objectives

The aim of the proposed research program is the development and experimental validation of a trajectory planning, management and control for a heterogeneous system of autonomous, cooperative vehicles. The proposed approach deals with analytical and application related aspects of levels of automation, information acquisition and data exchange (that is "intelligence"), so that each vehicle can participate to the whole mission or part of it optimizing its contribution to the betterment of the entire system.
In recent years, the general area of autonomous vehicles has received a considerable attention from the standpoint of basic research, applied research, and prototypical and production implementation. Development of "classical automation", that is control system, guidance, navigation, and path planning has greatly benefited by improved quality of sensors and actuators, their size reduction, by the increasing computation speed, and by reliable wireless communication and localization tools (GPS); thus allowing the use of more complex dynamic and kinematic control system methods. It has been possible to design autonomous systems ranging from mobile robots in a limited structured environment, to underwater vehicles for unmanned scientific exploration; from unmanned aerial vehicles for reconnaissance, search and rescue, to ground vehicles for de-mining and planetary exploration. Brilliant as the results are to the scientific world, as well as to the general public, still much needs to be done in order to classify it as a mature technology.
A step further in basic and applied research is the general problem of managing multiple autonomous vehicles (a formation, a swarm), which may be cooperating at various levels (mission, task, data exchange, sensor relay, etc.), up to the situation where the vehicles are heterogeneous in their nature, dynamic properties, performance, bandwidth and application area (underwater, land, air vehicles). The project objective is made up of several aspects, which may be integrated or sequential, and that relate to the path planning and motion coordination of a system of such vehicles (agents). They can be summarized as:
• Design and management of path planning methods (inclusive of control, guidance, and navigation) for each agent and their integration as an autonomous system,
• Definition and design of an appropriate sensor suite, sensor fusion synthesis, and data communication network, in order to ensure complete system "controllability and observability",
• Design of appropriate, real time fault tolerance mechanisms (detection, isolation, reconfiguration), in order to guarantee safety and mission success,
• Modular design of avionic and autronic components based on a "COTS" approach, and constraints on reliability, affordability and size,
• Development of a mission centre and experimental scenario, for the validation of technology.
The development of path planning and overall system manager will be based on combinatorial optimization and rule-based task selection. These techniques will be simplified to accept sub-optimal solution ("greedy") for online real time capability. Multiple tasks maybe assigned to all or a limited number of agents, depending on the problem in the initial phase of the mission. The multi-sensor system is the combination of all measurement components that is necessary to the complete control of a single agent, and functions as support to the other agents. To this end, the high level management system must include sensor fusion capabilities to allow sufficient decentralized as well as centralized information levels. The methods will have to take into account in fact information coming from proprioceptive sensors such as: inertial units, odometers, altimeters, gyros, as well as heteroceptive sensors like cameras, laser scanners, ultrasound and infrared device, A potential approach includes the improvement of deterministic models by including partially observable Markov processes techniques. Available information will be transmitted among agents via communication network (radio link, wireless and/or relayed by the mission center). To this end, safeguard against component failure and communication system "soft-failures" is critical. Fault tolerance is studied at the software level, since hardware redundancy is not feasible due to size, weight, and cost constraints. The fault tolerance system must perform diagnostic checks to sensors (sensor biases) and communication channels (packet loss), and replace faulty measurements or data flow with their best estimate. Methods used are the application of hybrid systems theory, combined with basis switching logic.
The methods outlined above, will be integrated with a laboratory and experimental setup, designed for validation purposes using a "COTS" approach. Modularity of avionics and on-board equipment will be highlighted by defining subsystems (flight management, telemetry, remote command, health monitoring, and mission management), and segments or tasks (flight segment, ground segment, identification segment, recon segment…).
The final aspect of the overall objective is the definition and test of a sample mission consisting of three agents (two aerial and a ground vehicle) operating with different but complementary tasks to achieve successful search and rescue. <<<

Timescale

24 months

National and international background

The research project has the aim of designing a management system for path planning, navigation and guidance of autonomous, cooperative heterogeneous vehicles. There are many foreseen uses for such systems that cover scientific applications, human and social aspects, defense and civilian protection inclusive of military and police tasks. In the sphere of interest of the present project we could mention as potential applications:
• Research and rescue operations in areas of morphologic asperities
• Reconnaissance and terrain mapping
• Surveillance operations in traffic congested areas and crowded areas
• Localization and relief operations after natural disasters such as earthquakes
• Cooperation to police and other authorities for border defense, terrorist attacks, handling and identification of hazardous material and mines.

In all these cases, the capability of using multiple autonomous cooperative vehicles in a closed loop fashion greatly increases the probability of success.
The proposed approach deals with analytical and experimental aspects for the integration of levels of automation, intelligent behavior, data exchange, so that each agent's dynamic properties are used optimally. The research team scientific capabilities are well integrated and complementary see [1], [2], [3], [4], [5], [6], they are documented in detail in the proposal written by each research unit, and some of the scientific personnel has direct recent experience in funded national projects [1) Progetto Nazionale MURST- MM09248181, 2001-2002 "Uso di Tecnologie Satellitari per il Controllo della Navigazione Aerea e Marina in Spazi vincolati", 2) Progetto Nazionale MIUR-2002098984, 2002-2004 "Sviluppo di sistemi integrati di guida, controllo e gestione di missione per il volo autonomo di velivoli non pilotati".
In recent years there has been a widespread interest in the planning and control of autonomous vehicles [7], [8]. This is due mainly to improved quality of sensors and actuators, their size reduction, to the increasing computation speed, and reliable wireless communication and localization tools (GPS); thus allowing the use of more complex dynamic and kinematic control system methods [9], [10], [11], [12]. Historically, among the first aerial implementations (UAV) we can mention the PREDATOR [14], [15], widely used for tactical reconnaissance. Civilian unmanned aircraft were also developed (CUAV), and an example is the CAMEX program [16]. Ground and marine examples are numerous as well [17], [18], [19], [20].
A step further in basic and applied research is the general problem of managing multiple autonomous vehicles (a formation, a swarm), which may be cooperating at various levels (mission, task, data exchange, sensor relay, etc.), up to the situation where the vehicles are heterogeneous in their nature, dynamic properties, performance, bandwidth and application area (underwater, land, air vehicles). Some similar challenges can be found in Air Traffic Control Management problems in a free flight environment [21], and management of mobile robots in constrained areas. The differences with the present research are many however and they refer to the type of vehicles, their interaction with the outside environment, and their sensitivity to reliable communications.
Within the proposed research we can identify four main areas: A) Path planning and control, B) Sensor design, communication network, and fault tolerance, C) Hardware components and mission center, D) Experimental phase.
A) Path planning and control
The management and path planning control of the system of interest is very complex. Among the various components, the inertial system (IN) has a very important role. In the case of map-based or landmark-based one of the tasks is the development of localization algorithms for the vehicles. This is called SLAM problem, which has been studied widely in different areas [22], [23]. The majority of the solution is based on estimation techniques integrating sensor information with vehicle's dynamics. If the landmark location is not known or partially known apriori, the problem becomes more complex, with a further difficulty if more vehicles are present within the scenario [24]. IN SLAM a critical aspect is the uncertainty level associated with the estimates. Kalman or Bayesian approaches are capable of providing confidence levels; however the presence of different vehicles requires the error to be greatly reduced. To this end, SLAM methods will be improved using a set membership approach [25], [26], [27], [28].
"Intelligent" planning among the vehicles is also critical. There are many approaches mostly limited to homogeneous vehicles' formations (mini satellites in Earth orbit, swarms of air vehicles, etc.) that range from MILP optimization, to game theory; from expert systems, to advanced nonlinear control techniques [29], [30], [31], [32], [33]. The state of the Art is much extended for aerospace applications, and there are several schools of thought worldwide. We can mention cooperative formation flight using optimization [34], task scheduling among different vehicles [35], [36], concentrate and distributed parameters methods, and methods based on architecture optimization [37].
Some of the challenges are relative to trajectory planning in the presence of moving obstacles, or obstacles' changing shape or their dynamic behavior, and a potential approach is the use of communication channel at different bandwidth, depending on the obstacle characteristics.

B) Sensor design, communication network, and fault tolerance
The capability of autonomous and cooperative mission planning, the capability of changing single tasks and level of information for path changes require a large number of sensors, online data exchange, and robustness to failures.
Sensor fusion is the primary tool for using the sensors in an optimal fashion, and significant results were obtained even in the case of multiple agents [38], [39]. The need of sensor fusion is also required to improve single vehicle motion capabilities, by adding additional information coming from selected sensors in other vehicles.
The vehicles are usually heterogeneous with different capabilities, but they must achieve their tasks in a co-operative and consistent way. Each vehicle, autonomously and incrementally, builds and executes its own plans taking into account the multi-agent context. The vehicles are able to know their positions with respect to the others, to collect the other vehicles plans and to coordinate their own plans with the other vehicles plans to produce "coordinated plans" that ensure their proper execution. There are many tasks for which distributed viewpoints are advantageous such as, surveillance, monitoring and de-mining. In these applications, sensor fusion is thought as sensor information exchanging among vehicles. In this context, it is possible to identify three main steps that can very often be developed separately: the decomposition of a mission into tasks, the allocation of the obtained tasks among the vehicles and the task achievement in a multi-agent context. In an unmanned vehicle the sensory systems are generally composed by a set of proprioceptive sensors for providing data on the evolution of internal variables (system state), and by a set of heteroceptive sensors for providing data for external variables (environment knowledge). The integration of sensory systems is a key tool for increasing the autonomy and reliability of these vehicles. In this framework reinforcement learning techniques are also analysed. For example the methods based on Partially Observable Markov Decision Process (POMPD) are used for supporting the localisation and navigation of autonomous agents.
A reliable communication network is also needed by the path planning manager. The research will be directed toward bi-directional links using radio modems or wireless protocols capable of high frequency exchange rate.
Information reliability needs the capability of performing fault detection, isolation, and accommodation procedures analytically, since hardware redundancy is not feasible [40]. Standard procedures are based on filter design generating residuals. Methods using parity space and observers are behind the mentioned approaches. Investigation will be performed on the use of geometric methods (Massoumnia), and Frisk's polynomial technique. From a methodological viewpoint, the problems cited above can also be treated uniformly making reference to the problem of effective control design for conditions of "REDUCED CONTROL AUTHORITY (RCA)".

C) Hardware components and mission center
On board component design, task scheduling, and a mission center are an integral part of the project. These aspects will be addressed using some of the facilities available at the proponents' laboratories.
The first aspect is based on previous and current experience gained with guidance, control and integrated inertial/GPS navigation design for the autonomous flight of a small aerial vehicle, and the experience gained in path planning for kart-like autonomous vehicles.High level task design is based on coarse grid of events such as approach, identification, localization, execution, verification. These tasks are programmed in each vehicle, but their order and number can vary online according to the scenario and task-specific performance indices. At the lower level a modular approach will be used based on subsystems and segments, an approach that has been successfully tested in previous research programs.
The level of development of a mission center and architecture for the project will depend on the level of funding devoted to the research. The structure of the mission center is divided into two parts: ground mission and motion mission. The ground mission is actually the control component. It will provide mission definition, mission monitoring, mission simulation, database, and it will relay to each vehicle the appropriate information, when intra-vehicular communication is not possible. The motion component of the mission center is actually the formation of heterogeneous vehicles including of their task allocation, trajectory planning, and data exchange with the other vehicles and the ground segment. The general architecture will consist of a decision level, and execution level, and a functional level.

D) Experimental Phase
The final aspect of the overall objective is the definition and test of a sample mission consisting of three agents (two aerial and a ground vehicle) operating with different but complementary tasks to achieve successful search and rescue operations. The proponents of the project will devote to this phase their hardware resources, however the success is depending on the funding level. <<<