ADVOCATE II - Jump into Intelligent Robotics

As Autonomous Vehicles are currently undergoing a transition from a research tool to real application, they are expected to work more reliably and safely. There are additional requirements for the design and operation of such vehicles in terms of fault-detection and -recovery and real autonomy (adaptability to unforeseen situations).

The aim of ADVOCATE II is to design and develop an architecture to increase the performance of unmanned underwater and ground robotic applications.

ADVOCATE is the abbreviation for ADVanced Onboard diagnosis and Control of Autonomous sysTEms.

The objectives are to increase the safety for the system itself as well as the environment, to increase automation and to increase efficiency and reliability of the system. These objectives will be reached by adding intelligence into existing and new control software to diagnose and recover from any dysfunction situation of the system. The architecture is designed with the ability to incorporate and merge different AI techniques.

The main objective is to have a better management of uncertainty in robots by the use of intelligent diagnosis and control software, but without too specific non-reusable developments.

End-Users

Three end-users are involved in the ADVOCATE II project:

• Ifremer (France) designing ROVs (Remotely Operated Vehicles) for scientific applications
• ATLAS Elektronik (Germany) designing AUVs (Autonomous Underwater Vehicles) and semi-AUVs for industrial applications
• University of Alcalá de Henares (Spain) designing Piloting Modules for either Autonomous or Remotely Operated Groung Vehicles (AGVs or ROGVs, respectively) for surveillance applications.

End-User Applications

Several diagnosis problems are considered for each end-user, involving different kinds of failures:

• Thruster or motor failure diagnosis and recovery (Ifremer, ATLAS and UAH). As soon as an abnormal behaviour of the vehicle is detected, the system provides a diagnosis on the responsible thruster or motor, and a recovery action is issued based on the redistribution of propulsion power.
• Sensor Malfunction (UAH and ATLAS). Diagnosis on sensors state is provided so as to account for failure situations, or in case information coming from sensors is corrupted by acoustic noise or interferences.
• Battery monitoring (UAH and Ifremer). Both an AGV and an AUV are supplied with energy by their own battery. An Intelligent module is in charge of managing the mission parameters according to the power consumption in order to avoid inopportune mission abortion.
• Abnormal global behaviour (ATLAS). An Intelligent Module will be developed to provide monitoring and assessment of the motion characteristics and the control inputs of the vehicle.

End-User Vehicles

Ifremer Vehicle

The Ifremer vehicle is based on an experimental underwater vehicle, called VORTEX, operated in a test pool or in simulation. The vehicle is a small experimental Remotely Operated Vehicle (ROV), but it can be considered as an AUV from the control point of view since fully automatic missions can be programmed and performed. VORTEX can be considered as a AUV, not dedictated to long survey tasks, but to intervention tasks (offshore application, for instance). Mechanically, the vehicle structure consists of a basis tubular structure on which different actuators are arranged, without pre-defined locations, as shown in the figure. Central to this structure is the main electronics package containing the vehicle electronics as well as the different set of sensors.

ATLAS Vehicle

The DeepC vehicle developed under the support and promotion of the Federal Ministry of Education and Research of Germany is a fully autonomous underwater vehicle with the related components on the water's surface for oceanographic and oceanlogic applications as shown in the figure. One of the outstanding features of the DeepC is the reactive autonomy. This property allows situation-adapted mission and vehicle control on the basis of multi-sensor data fusion, image evaluation and higher-level decision techniques. The aim of the active and reactive process is to achieve high levels of reliability and safety for longer underwater missions in different sea areas and in the presence of different ground topologies.

UAH Vehicle

In the context of the ADVOCATE II project, UAH deploys a ground vehicle that works in a combination of autonomous and teleoperated mode. The vehicle is intended to perform surveillance tasks after hours in a large building composed of corridors, halls, offices, laboratories, etc. For this purpose, UAH is currently deploying the BART robot shown in the figure. The operator is in charge of global vehicle navigation by remotely commanding its actuators according to the images that are continuously transmitted through a wireless ethernet link from the vehicle to the base station.

The ADVOCATE II architecture

The ADVOCATE II architecture introduces intelligent techniques for diagnosis, recovery, and re-planning into different types of robotic applications. The global objective of the project is to enhance the level of reliability and efficiency of autonomous robotic systems, as described above by:

• Constructing an open, modular, and generic software architecture for diagnosis and control of autonomous robotic systems.
• Developing or improving a set of intelligent diagnosis modules fully compatible with this architecture and tested in operational applications.
• Carrying out practical tests and demonstrations on a set of operational prototypes in order to prove operational usage and efficiency of this solution in several application fields, and particularly for Autonomous Underwater Vehicles and Autonomous Ground Vehicles.

The ADVOCATE II architecture is a distributed architecture, which is based on a generic communication protocol between the different modules. The architecture is modular and easy to evolve and adapt to legacy piloting systems. It comprises five different types of modules, which are described below and possibly a number of Man Machine Interfaces (MMIs). The architecture is designed to allow easy integration of different artificial intelligence techniques into preexisting solutions.

The ADVOCATE II architecture is organised around the Directory module, the central point for communication between modules. The technology used is based on XML and some innovative recent technologies, as SOAP (Simple Object Access Protocol) and UDDI, upon HTTP communication protocol. HTTP protocol is lightweight in itself. Simple data packages (small XML documents) are sent and are easy to control by modules, in order to limit overload.

The ADVOCATE Modules

Robot Piloting Module (RPM)

This module manages the mission plans and communicates directly with the vehicle sensors and actuators. Several RPMs, each of them working on a specific subsystem, can be plugged on the ADVOCATE II architecture. Each end-user participating in the project (UAH, ATLAS, and Ifremer) is responsible for the corresponding piloting modules.

Decision Module (DM)

This module has a generic part and a specific part containing knowledge for decision making, according to diagnosis results. The DM manages the overall diagnosis and recovery process, including the following functionalities:

- control of the monitoring/diagnosis/recovery process,

- integration of uncertainty information provided by the intelligent modules,

- validation of diagnosis and recovery actions (if needed),

- interaction with human operators (if any) with regards to diagnosis and recovery.

- conversion of the recovery actions into recovery plans.

Intelligent Modules

Several Intelligent Modules for each application are currently being developed, using different Artificial Intelligent techniques devoted to solve real problems on operational robots by making use of specific knowledge on them. Intelligent Modules include functionalities providing a diagnosis (identification of sub-system state), a proposed recovery action, or both. The present implementation comprises modules based on:

- Bayesian Belief Networks (BBN).

- Fuzzy Logic (FL).

- Neuro-Symbolic Systems (NSS).

Directory Module

The Directory Module is a central point of the architecture. It is intended to be implemented using a Java UDDI tool supplied by IBM. The Directory Module will allow to progressively integrate all the intelligent and piloting modules of the ADVOCATE II project. This objective compounds the design of the upgrading features to add to the SOAP implementations, in order to integrate the soft real time specifications.

Configuration Tool

It is an offline friendly application which eases the production of the XML Configuration File for every modules of the ADVOCATE II system. By generating a graphical view of the system the user will be able to check the concordance of the configuration files, and to foresee the behaviour of the modules in the system.

Artificial Intelligence Technologies Used

Bayesian Networks (BN)

The BN is a model representing the causal relations between the entities of the modelled domain. An influence diagram adds decisions and value functions to the model. The strengths of the relations are described using probabilities. Utility functions describe the preferences of the decision-maker. BNs and IDs can be adapted to many classification (diagnosis) or decision problems, particularly in case of erroneous, incomplete or uncertain data, or problems that involve sensitivity analysis, conflict analysis, or calculation of value of information.

Neuro-symbolic system (NSS)

An incremental neuro-symbolic system (INSS) represents the initial expertise of the domain as symbolic rules written by the experts. These rules are compiled into a neural structure to be used during the on-line diagnosis. The compiled neural network is trained and tested on a set of representative examples. The refinement of the neural network can be performed when badly classified examples are encountered during the system functioning. These new examples are then added to the initial learning base. The knowledge of the system is then increased and the conservation of the initial knowledge is guaranteed.

Fuzzy Logic (FL)

A Fuzzy system represents (symbolic) expert knowledge by means of fuzzy rules. Fuzzy rules use linguistic variables (which values are linguistic labels) to describe a decision or control protocol in terms that are quite close to the language used by the experts. That "proximity" between the language used by the experts and that representing the fuzzy rules simplifies the process of knowledge extraction, and makes the decision process understandable by the experts. In addition, the underlying reasoning methods are particularly well adapted to decision or control problems working with uncertain or noisy data.

In conclusion

The main objective of the ADVOCATE II project is to develop a software architecture to allow the implementation of intelligent control modules for underwater and ground robotic applications, in order to increase their reliability.

The interest of such a concept from the marketing point of view has been demonstrated by a market study. Additional ongoing information concerning the ADVOCATE II project can be found at the project web site: http://www.advocate-2.com .

Scientific Articles

• Kjærulff, U. B. and Madsen, A. L. (2004), A Methodology for Acquiring Qualitative Knowledge for Probabilistic Graphical Models, Proceedings of the International Conference on Informational Processing and Management of Uncertainty in knowledge-based Systems, pages 143-150.
• Kalwa, J. and Madsen, A. L. (2004), Sonar Image Quality Assessment for an Autonomous Underwater Vehicle, Proceedings of the 10th International Symposium on Robotics and Applications.
• Madsen A. L., Kjærulff, U.B., Kalwa, J., Perrier, M. and Sotelo, M. A. (2004), Applications of Probabilistic Graphical Models to Diagnosis and Control of Autonomous Vehicles, Proceedings of the second Bayesian Application Modeling Workshop.
• Sotelo, M. A., Bergasa, L. M., Flores, R., Ocana, M., Doussin, M-H., Magdalena, L., Kalwa, J., Madsen, A. L., Perrier, M., Roland, D. and Corigliano, P., (2003), ADVanced On-Board Diagnosis and Control of Autonomous Systems II, Computer Aided Systems Theory --- EUROCAST 2003, Springer Verlag Lecture Notes on Computer Science, 2809, pages: 302-313.