Fault-Tolerant Group Communication forCooperative Mobile Robots Xavier D´efago School of Information Science, Japan Advanced Institute of Science and Technology, 1-1 Asahidai, Nomi, Ishi
Trang 1Fault-Tolerant Group Communication for
Cooperative Mobile Robots
Xavier D´efago
School of Information Science, Japan Advanced Institute of Science and Technology,
1-1 Asahidai, Nomi, Ishikawa 923-1292, Japan, Email: defago@jaist.ac.jp, URL: http://www.jaist.ac.jp/~defago/
1 Context and Illustration
More and more projected applications involving autonomous mobile systems (e.g., robots, smart vehicles, MEMS, or mobile sensors) rely on the ability for several nodes to cooperate toward a common goal For instance, search and rescue, intelligent transport, or unmanned space exploitation all require several autonomous nodes to act as a coherent group We focus on the mechanisms for cooperation among mobile robots and consider a semi-autonomous gardening system as a concrete illustration to the problem
The gardening system could be described as follows The robots, disguised
as small critters to avoid attracting attention, constantly monitor the garden and perform various small maintainance tasks as required Among other things, these tasks include removing dead leaves, pruning trees, watering and weeding the garden, and providing additional nutriments where needed Simple tasks, such as weeding, require the attention of a single robot and can be decided locally Other tasks, such as giving nutriments or pruning, require that other robots be kept properly informed Some tasks, such as getting rid of a dead tree, may even require the coordination of many robots since one single robot would
be unable to carry it out alone
2 Problem Statement
The system described above is inherently very dynamic For instance, require-ments may change significantly with time, robots may become deffective in sev-eral ways, new ones may be added, extreme environmental conditions may re-quire compensation, or unexpected events may induce additional activities In other words, this means first that the system as a whole must have the capacity
to adapt to frequently changing conditions However, it is difficult to anticipate all problems Consequently, the system must rely on mechanisms that allow it
to exhibit a correct behavior, even after unexpecte changed In other words, it requires mechanisms to be self-stabilizing
A second issue is that many different tasks require that several robots ro-bustely synchrnize, coordinate their actions and behave in a consistent way
Trang 2Achieving such coordination is definitely not an easy task and requires to rely
on complex communication and coordination protocols This raises one impor-tant question which is how to effectively program the behavior of robot groups rather than individual entities
3 Groups vs Swarms
Many researchers try to solve the problem by trying to build robot swarms
in which simple individual actions can combine to complex group behaviors The idea is to learn from the mechanisms involved in the behavior of social insects such as ants or bees The approach is very interesting from an intellectual viewpoint, and can potentially address the two issues raised before
However, although swarm behavior and artificial life may be very effetive from the viewpoint of the robots, it ignores one important part of the picture; the fact that requirements must be met in a predictable manner and that the system must remain controllable by its legitimate users
For this last reason, we pursue a different approach to the problem, where
we try to rely on protocols and mechanisms of which we can establish the cor-rectness In particular, we try to keep the link with existing research on fault-tolerant distributed protocols and their application to mobile robot systems In distributed systems research, agreement and coordination is usually achieved through group communication
In the presentation, we will describe different ways in which we try to address the problem On the one side, we look at group communication and distributed mutual exclusion as a basis for the coordination of robot movements Unlike other similar work, we consider that the timing characteristics of the system are unpredictable
On the other side, we consider simpler models to look at more fundamental questions related to the limitation of reaching an agreement between robots For instance, some very simple problems, such as gathering robots at the same location, cannot be solved in a deterministic way The use of other devices, such
as a compass, or randomization, can provide a solution to the problem