Date: Fri, 21 Jan 2000 17:08:17 +0800 From: Thomas Braunl To: pere@ee.uwa.edu.au Hi Petter here is the text passage from Klaus and I put in the headlines for the requested sections. Since we want funding for 3 Ph.D.s I suggest the project structure: - robot operating system - robot simulator - behaviour-based application What do you think? Thanks Thomas P.S. can you handle MS Word files as well? ________________________________________________________________________ Thomas Bräunl e-mail braunl@ee.uwa.edu.au Dept. of Electrical & Electronic Engineering phone +61 8 9380-1763 The University of Western Australia fax +61 8 9380-1168 Nedlands, Perth, WA 6907, Australia http://www.ee.uwa.edu.au/~braunl ________________________________________________________________________ 17.1 Aims, significance and expected outcomes Aims In the broad field of robotics and especially in the field of the Autonomous Mobile Robots (AMR) the subject of service robotics has received growing attention in recent years. Service robots need to be intelligent and be easy to handle by the growing number of non-professional operators. To fulfil this demanding task, alternative solutions to the traditional approaches have to be found. The classic approaches use the well-known Artificial Intelligence (AI) planning methods to solve a given problem [LATOMBE 91]. They can only cope with situations which the programmer has foreseen and implemented in the robot's control logic. The robot will be helpless if something unpredictable happens so that the pre-calculated actions cannot be performed. The next step in this development was the biologically inspired behaviour-based approach [BROOKS 86]. It is based upon a set of simple behaviours which are independent of each other. Each individual robot behaviour consists of coupling sensor inputs to actuator outputs to produce a desired action. The behaviours are organised in priority order. A so called arbiter chooses the action with the highest priority and therefore controls the total behaviour of the robot which makes it appear to be intelligent. This approach has proven to be robust in the real world with minimal environment modeling. However, it has a limited capability to learn or acquire knowledge and has few mechanisms to detect failure. Consider the example of such a robot detecting an obstacle while aiming for a certain goal in its environment. The system can become deadlocked between competing behaviours i.e. escaping the obstacle and approaching the goal. Several simple techniques have been introduced to solve this problem without resorting to planning, i.e. wandering behaviours [ANDERSON90] or relocating the goal [ADAMS90]. Nevertheless, the robot has learnt nothing from its ordeal and will behave in the same inefficient manner if it encounters the same obstacle again. This shortcoming has led to the development of hybrid planning-reactive systems [ARKIN90]. The robot monitors the outcome of the response to the sensor stimulus and decides if another response needs to be coupled to this sensor input. In other words, to operate efficiently a behaviour-based robot must be able to learn to co-ordinate its behaviours. The aspect of machine learning is a very interesting and up to date topic where research is in its infancy. A very promising and popular approach is the use of so called mobile agents [FRANKLIN96]. The term "agent" reflects a change of view towards programs; an "agent" has to be able to cope with broad user expectations. This new view also has new requirements for machine intelligence, i.e. a user can give the robot the same orders as he would give to a human. An agent is a program that acts and decides autonomously in the execution of an order. In contrast to this, a traditional program is considered to be passive, as it simply reacts on a user input with a certain output. To give an example, imagine a hospital with an autonomous mobile robot that can bring medicine to patients. The doctor tells the robot to bring a certain medicine to patient X in room Y. The AMR knows where it can get the medicine and drives to the distribution point. There it contacts a computer to automatically hand out the medicine. Bad luck, this specific medicine is not in stock. So the AMR contacts a medicine database to look for an equivalent product with the same effect which it can take to the patient. Now just in front of room Y it discovers that patient X is no longer lying here. A search through the hospital's database reveals that patient X moved to room Z where the AMR redirects itself to. It can easily be seen that classic approaches would not have been able to solve such a complex task. The designer of the mobile agent has of course trained the robot to perform many different behaviours, but the agent makes the decisions on its own and progressively learns new optimal behaviours. This concept of mobile agents can be extended to a group of agents. Members of this group can exchange information to cooperate or coordinate the next steps in the same task. This requires methods of communication and some kind of social behaviour, which brings us to a form of artificial life. A classic example of a task for such a group of robots is the exploration of an unknown environment. Each member of the group explores a section on its own and upon return to a meeting point merges its information with the information of the other members to build up a complete map. Another example is the transportation of an unwieldy object. The robots have to co-ordinate their actions to find suitable positions around the object to move it together. There are widespread applications for an intelligent group of robots which have not yet been researched. Most of the ideas and techniques derive from biological models and are implemented by using the concepts of Artificial Intelligence. To ease implementation and testing of control programs for mobile agents, we intend to build a modular and distributed robot simulator. Agents interacting in a real-world environment have some similarities to asyncron distributed systems, and it seems obvious that a distributed system is the best way to simulate this aspect of the problem domain. The simulator will be modular to handle new sensors without recompiling. The sensor modules will be loaded based on a robot description file which list the available sensors and the placement on the robot. We intent do build modules distance sensors, radio communication, compass sensor and simulated camera. The camera simulation will use freely awailable 3D tools to generate the robot view in real time. The simulator will be open source and free to use for any other research project, and we intend to use only freely available libraries to make sure this system will work on all platforms. In 1992, Oliver Faugeras published a paper describing a method for automatic selv-calibration of moving cameras or stereo vision. We intend to implement this method on low resolution cameras and low CPU power equipment. If we manage to solve the following equation with 9 unknown coefficients in real time, our solution should give very high frame rate on high CPU computers. ... XXX fill in equation Significance xxxx Expected outcomes xxxx The outcome of the project will be as follows: - xx, - xx, - Simulation system. 17.2 Research plan, methods, techniques Research plan xx. Methods and techniques xx 17.3 Proposed timing xx (diagram). January 2001 to March 2001 xx April 2001 to June 2001 xx Robot Operating System April 2001 to October 2001 xx November 2001 to March 2002 xx April 2002 to March 2003 xx Robot Simulator April 2001 to September 2001 xx October 2001 to March 2002 xx April 2002 to March 2003 xx Behaviour-Based System January 2002 to December 2002 xx January 2003 to July 2003 xx Evaluation and Improvement June 2003 to December 2003 Evaluation and Testing October 2003 to December 2003 Writing of final report 17.4 Justification of the budget xx. We need three Ph.D. students to work on this project because ... 17.5 Relevance of applicant skills, training and experience to the project At Stuttgart Univ., A/Prof. Thomas Bräunl cofounded the "Robotics Lab" with Prof. Paul Levi. This lab conprised three industrial size mobile platforms, one carrying an industrial robot manipulator arm, one carrying an innovative stereo head system. These robot shave been used in several research projects, including: - Implementing a robot operating system in the language Oberon - Tracking and following a leading vehicle - Object recognition and flexible manipulation - Cooperation of multiple vehicles in docking and transportation tasks - Object tracking with the help of a massively parallel stereo camera system
At UWA, A/Prof. Thomas Bräunl founded the "Mobile Robot Lab", which comprises currecntly 15 mobile robot systems of the "EyeBot" family. These are non-holonomic robots with 2 wheels, holonomic 4-wheel-driven robots, 6-legged walking robots, and 2-legged walking robots.
17.6 Role of each participant The chief investigator xxx. The three Ph.D. students will ... References [ADAMS90] MD Adams, H HU and PJ Probert, "Towards a Real-Time Architecture for Obstacle Avoidance and Path Planning in Mobile Robots", Proceedings of IEEE on Robotics and Automation Conference, pp584-589, 1990. [ARKIN90] RC Arkin, "Integrating Behavioural, Perceptual and World Knowledge in Reactive Navigation", Robotics and Autonomous Systems, vol.6 pp105-122, 1990. [ANDERSON90] TL Anderson and M Donath, "Animal Behaviour as a Paradigm for Developing Robot Autonomy", Robotics and Autonomous Systems, vol.6 pp146-168 1990. [BROOKS 86] RA Brooks, "A layered Intelligent Control System for a Mobile Robot", IEEE Trans. On Robotics and Automation, RA-2, April 1986. [FRANKLIN96] S Franklin and A Graesser, "Is it an Agent, or just a Program: A Taxonomy for Autononmous Agents.", Proc.3 Int. Workshop on Agent Theories, Architecture, and Languages (ATAL), pp 193-206, Budapest 1996 [LATOMBE 91] J-C Latombe, "Robot Motion Planning", Kluwer Academic Press, 1991. <>