Mobile robot navigation using motor schema and fuzzy context dependent behavior modulation

Abstract

This paper presents a novel technique to autonomously select different motor schemas using fuzzy context dependant blending of robot behaviors for navigation. First, a set of motor schemas is formed as behaviors. Both strategic and reactive type schemas have been employed in order to facilitate both the aspects of global and local motion planning. While strategic schemas are formed using the prior knowledge of the environment, the reactive schemas are activated using current sensory data of the robot. For global path planning, a safe path is first created using a Voronoi diagram. For local planning, the Voronoi vertices are treated as immediate subgoals and are used to form schemas leading to achieve optimized traveled distance and goal oriented robot navigation. Two motor schemas are formed as reactive behaviors for obstacle avoidance. The unknown obstacles are modeled using the sensory data. The coordinated behavior is achieved while employing weighed vector summation of the schemas. The adaptation of weights are achieved through a fuzzy inference system where fuzzy rules are used to dynamically generate the weights during navigation. A novel approach is proposed for fuzzy context-dependent blending of schemas. Fuzzy rules are formed using two main criteria into account: the first criterion reasons out the context dependent activity of a schema for achieving goal and the second criterion reasons out cooperative activity of strategic schemas with high priority reactive schemas. Comprehensive results validate that the proposed technique eliminates the existing drawbacks of motor schema approaches available in literature and provides collision free goal oriented robot navigation.

Introduction

Development of an intelligent control system becomes a challenging issue in mobile robot navigation, where an appropriate control action is executed according to uncertainties associated within sensory systems and the surrounding dynamic environment. Currently, there exist three broad categories of intelligent architectures: centralized (deliberative), behavior-based (reactive), and hybrid (deliberative-reactive). Centralized architectures [1], [2] create a complete model of the static environment by combining all available sensory data. Then it employs deliberative planning in order to generate a series of actions within the context of the static model to accomplish a given task. After successful execution of an action, the robot stops, gathers more information, and repeats the process. An important aspect of this architecture is the top-down approach of planning, where high level constraints are integrated into low level control commands. Centralized intelligent controller can coordinate multiple goals and constraints within a complex environment. However, planning a series of actions by sensor fusion in a centralized architecture introduces a potentially harmful delay [3]. Moreover, the system may fail entirely if any single part fails, e.g., sensor fusion or planning is not functioning properly. This leads to inappropriate use of centralized controller for a real time system, where the environment is dynamic or uncertain. On the other hand, behavior-based architectures [4], [5] are composed of independent task-achieving modules, or behaviors. Each behavior receives a particular sensory perception which is related to a given task, thus avoiding the need for sensor fusion. Moreover, task-achieving aspect of each behavior leads to distributive control process, which reduces planning complexity and increases responsiveness to a dynamic environment. Behavior-based architectures possess bottom-up approach of decision-making as they do not integrate high level constraints in action generation process. They are also more robust because if any behavioral unit of the system fails, the other units continue to function independently. However, a completely distributed system does not reflect the multiple objectives and constraints that the system is subjected to at any given moment, thus leading to significantly suboptimal performance [3], [6] and unreliable decision-making [7]. Additionally, the interactions, both between the behavioral units and between the system and its environment, are less predictable and more difficult to understand and modify as compared to a purely centralized system. Hybrid architectures [8], [9] attempt to combine the goal-directed solution of centralized architecture and responsiveness of behavior-based architecture. The top level of a hybrid architecture is a deliberative planner, which assimilates all available information and creates long-term global plans. The lowest level consists of a behavior-based architecture, which recommends real-time responses to sensory stimuli. The most important part of a hybrid architecture is the intermediate level, called behavior coordinator or sequencer [10]. A behavior coordinator takes into account the high level constraints of a deliberative planner and real-time responses of individual behaviors. As a result, it can generate an action, which satisfies the objectives of both the planner and behaviors.

This work attempts to employ a fuzzy logic based behavior coordination system to achieve robust performance in mobile robot navigation. Fuzzy rules are used for combining high level constraints of a deliberative planner with realtime responses of individual behaviors to generate an action, which satisfies the objectives of both the planner and behaviors.