A Reactive Anticipation for Autonomous Robot Navigation
Emna Ayari, Sameh El, Khaled Ghedir · InTech eBooks · 2012
Nowadays, mobile robots are expected to carry out various tasks in all kinds of application fields ranging from manufacturing plants, transportation, nursing service, resource or underwater exploration. In all these applications, robots should navigate autonomously in uncertain and dynamic environments in order to achieve their goals. So, the most current challenge in the development of autonomous robot control systems is making them respond intelligently to changing environments. Navigation in such environments involves many mechanisms such as: object detection, perception, internal building model, decision making, prediction of the future state of the environment and on-line navigation. To attend its goal safely, the robot should minimise interaction with other actors in order to avoid conflict situations. Generally, this problem comes up when many robots and/or actors would have access to the same space at the same time. In this case, the control of autonomous robotic navigation for conflict resolution has been widely studied. Some researchers have been focused on navigation in dynamic environments, where either reactive systems (producing real time behaviour), deliberative systems (introduce reasoning and need much more time to calculate a suitable decision) or hybrid systems (combine deliberative and reactive approaches) have been used in order to attend a known goal. Typically, reactive systems are used to deal with simple problems (detect an obstacle, go away from an obstacle, follow a wall, etc.). Nevertheless, reactive systems are typically less affected by errors and do not require an explicit model of the environment in order to navigate inside an unknown space. Furthermore, they usually deal only with local information that may be captured at real time. However, in reactive systems, the robot can be derived to a conflict situation with other actors because they ignore prediction and reasoning in the decision process. To face this problem, global planning approaches are used. They consist of elaborating a global plan from beginning state to goal state. These approaches need prior complete information about the state of the environment, so they do not take into account the environment uncertainty. So, in more complex situations, hybrid approaches are used. They combine reactive approaches according to a higher level in order to include anticipation of the state of the environment in the decision process. In these cases, low level control operates in a reactive way (local navigation) whereas high level systems tend to be deliberative. It provides, at each step, a partial moving plan to the robot. But these systems are complex because they need much more time to calculate or to update a suitable trajectory toward a predefined goal. In this situation, it is interesting to introduce prediction