Behavior based planning method

Chemical signal guided search is a complicated problem. One way to reduce the complexity is to break down the planning problem into a set of simpler subproblems each solvable by simpler actions with an appropriate method to switch between actions. This divide-and-conquer strategy is effective in many planning applications that deal with complex systems. These simpler actions are called behaviors. A behavior is a mapping of sensor inputs to a pattern of motor actions, that accomplishes a single goal within a restricted context. A behavior-based planning (BBP) strategy is an efficient means to navigate an autonomous system in an uncertain environment. To use a set of behaviors to achieve a task a mechanism for coordinating the behaviors is also required.

In the late 1970's and early 1980's, Arbib began to investigate models of animal intelligence from the biological and cognitive sciences point-of-view to gain alternative insight into the design of advanced robotic capabilities (Arbib, 1981). At nearly the same time, Braiten Berg studied methods by which machine intelligence could be evolved by using sensor-motor pairs to design vehicle systems (Braintenberg, 1984). Later, a new generation of AI researchers began exploring the biological sciences in search of new organizing principles and methods of obtaining intelligence. This research resulted in the reactive behavior-based approaches. Brooks' subsumption architecture is the most influential of the purely reactive paradigms. Its basic idea is to describe a complex task by several behaviors, each with simple features (Brooks, 1986). Design of a behavior-based planner includes two significant steps. First, the designer must formulate each reactive behavior quantitatively and implement the behavior as an algorithm. Second, the designer must define and implement a methodology for coordinating the possibly conflicting commands from the different behaviors to achieve good mission performance.

Various coordination approaches have been proposed. For example, each behavior can output a command and a priority. Traditional binary logic can be used to select and output the command with the highest priority. An alternative coordination approach is to use artificial potential fields (Arkin & Murphy 1990). A drawback to either approach is that formulating and coordinating the reactive behaviors requires significant pre-mission simulation and testing. These are ad-hoc processes and may need to be re-addressed each time new behaviors are added or existing behaviors are changed. In some applications, these tuning parameters depend heavily on environmental conditions. Another alternative that has been suggested is to train an artificial neural network (ANN) to perform the behavior coordination (Li et al., 1997). However, this approach would require some mechanism for determining "correct" coordination decisions for each training scenario and would provide no guarantee that all coordination situations are properly trained (Berns et al., 1991). Fuzzy control methods can improve the performance of reactive behavior coordination (Li et al. 1997) by providing a formalism for automatically interpolating between alternative behaviors. Although similar in overall structure, fuzzy control differs from classic feedback control. In fuzzy control, the controller has the same function inputs and outputs as in the feedback control, but internally the control values are computed using techniques from fuzzy logic. Fuzzy controller takes fuzzy state variables, by applying sets of fuzzy rules, produces a set of fuzzy control values. These fuzzy control values are not precise numbers, but rather represent a range of possible values with different weights. Eventually, a decision is made based on the fuzzy control values.

Behavior based design methodologies are bottom-up approaches to the design of an intelligent system. Observed behaviors with simple features are analyzed and synthesized independently. By using these design methodologies, we break down the complicated plume tracking problem into five behaviors. Later in this chapter, we will describe the behaviors and coordination mechanism that were used to solve the problem of chemical plume tracing strategy for an AUV in details. The behaviors were inspired by behaviors observed in biological entities.

Learn Photoshop Now

Learn Photoshop Now

This first volume will guide you through the basics of Photoshop. Well start at the beginning and slowly be working our way through to the more advanced stuff but dont worry its all aimed at the total newbie.

Get My Free Ebook

Post a comment