Hybrid Navigation Control for Multi-Behaviour Robots
PRESENTER:
Fadi Halal Universite du Quebec a Outaouais
ABSTRACT:
This presentation addresses the issue of designing integrated
deliberative-reactive architectures for multi-behaviour robot navigation
control. The objective of the study is to devise and investigate a
methodology for designing robust planning and control systems equipped
with a high level of intelligence and capable of navigating a mobile
platform, at a high level of performance, in partly unknown environments,
where the mobile robot multi-task operation is subject to different
behaviours. Of particular interest in this research are
deliberative-reactive navigation systems operating in large, complex
environments, such as those applied in environmental monitoring, that make
use of a variety of remote sensing data. The spatial data are interpreted
intelligently using multi-layer feature maps.
A formal model for hybrid mobile robot navigation is presented. The model
integrates two levels of navigation, deliberative and reactive. The
novelty in this model is that the decision component makes a decision
depending on the global and local context choosing the suitable behaviour,
including conflicting behaviours, and regulates the relation between the
deliberative and the reactive navigation via computational intelligence
techniques. The presented methodology offers a suitable solution for
complex partially known environments, where the mobile robot control
produces an overall behaviour for executing the proper action in order to
reach the target by employing multi-task navigation. In terms of the
multi-behaviour operation, the following behaviours are considered:
different tasks for environment data acquisition, and different navigation
behaviours. In the latter type of behaviours, three situations are
studied: dynamic local path, unreachable local path, and conflicting
behaviours in a critical situation.
The experiments presented demonstrate the utility of the model in the two
fundamental types of the navigation: those with the predominance of the
deliberative action, and those with dominant reactive action. The
experiments adopted the following methodological approach. First, the
deliberative navigation was developed using a hybrid genetic algorithm to
deal with multi-task navigation. The proposed navigation system has a
flexible and efficient performance along global and local paths. A
complete solution for monitoring of water quality in Lake Winnipeg using
satellite data was presented. Second, a multi-behaviour
deliberative-reactive navigation scheme was designed to deal with
conflicting behaviours using artificial intelligence methods, such as
fuzzy systems and genetic algorithms, in a hierarchical configuration. The
fuzzy control drives the robot to execute the required behaviour,
depending on the robot-specific situation and the characteristics of the
environment, in order to reach a given target.