Workshop Report

IJCAI-99 Workshop Rob-3
Reasoning with Uncertainty in Robot Navigation

1. Workshop home page

http://aass.oru.se/Living/RUR99/

2. Topics

Dealing with uncertainty constitutes the focus of a large research effort in AI, which has led to the development of a number of new theories and new techniques. In the field of robot navigation no such research tradition exists, but recently several of the new approaches to uncertainty have been used to address the issues of uncertainty in robot navigation. The purpose of this workshop was to bring together researchers in the fields of robotics and of uncertainty in AI to discuss advanced methods of dealing with sensor and movement uncertainty in mobile robots.

The workshop was a follow up of the three day workshop ``Reasoning with Uncertainty in Robotics'' (RUR-95, Amsterdam, NL, Dec. 1995). The great interest raised by RUR-95 prompted us to organize a second RUR workshop to collect the new developments in both fields. A related aim of the workshop was to provide an opportunity to critically examine various (competing) approaches, and to discuss their strengths and weaknesses in several types of environments.

3. Attendance

27 persons registered before the workshop, but actual attendance was between 35 and 40, almost beyond the capacity of the room. Many of the last minute additions were people who had attended the other two IJCAI-99 workshops on robotics - see below. From the feedback received, attendants were pleased by the contents and the format of the workshop.

4. Progress made

The workshop consisted in the presentation of 6 contributed papers, 2 invited talks, and a panel discussion. The contributed papers were of high quality, and the invited talks were much appreciated by the audience. The workshop also included a ramp session in which each participant coud give a flash presentation of his or her work. Eight people used this spot. The ramp session was generally felt as a useful addition to the technical presentations. The workshop proceedings are available on-line at the workshop home page.

One of the objectives of the workshop was to assess the relative merits and demerits in the robotic field of the different formalisms for managing uncertainty that have developed in the AI community. This was the aim of the panel discussion. This objective has not been satisfactorily achieved, mainly because the non-probabilistic formalisms (e.g., fuzzy logic, possibility theory, and evidence theory) were under-represented at the workshop.

While the problem of a fair and useful comparison between different approaches to uncertainty remains unsolved, the workshop increased the general awereness of the necessity to look into different uncertainty formalisms, and to assess the adequacy of each formalism for each robotic task. It was felt that the issue of understanding the intended semantics of the uncertainty representations that we use in our robots cannot be waived any longer by just saying that ``it works''. This important issue will have to be addressed by future workshops.

In summary, the workshop was fully succesful in its being a forum for scientific exchange on the target issue, both in quantitative and qualitative terms. It was only partially so, however, in its goal of reaching a better understanding of the role of different uncertainty formalisms in our field. The awareness of the importance of this goal, however, has increased through the discussions at the workshop.

5. Related workshops

There were two more robotics workshops at IJCAI-99: Robot Action Planning (Rob1), and Adaptive Spatial Representations of Dynamic Environments (Rob2). The three workshops had much intersection, both in attendance and topics: the issue of uncertainty was strongly present in both Rob1 and Rob2; and several papers at Rob3 dealt with uncertainty in environment modeling. These overlaps confirm that uncertainty is the object of growing interest in the robotics community. In fact, the community might respond well to a (possibly larger) workshop on uncertainty in robotics that explicitly include the sub-topics of decision and planning under uncertainty, and of mapping and self-localization under uncertainty.

6. Follow-up

As mobile robots are going to be used for increasingly complex tasks, the issue of dealing with uncertainty will keep cropping up. The participation at this workshop shows that there is a need for a continued interaction between the uncertainty in AI and the robotics communities on this issue. Moreover, there is a growing interest in the yet-to-be-answered question of how to match uncertainty formalisms to robotic problems. All this suggests a need for a follow-up RUR workshop, and even for a series of RUR workshops on a regular basis.