31-08-2012, 10:32 AM
Neurofuzzy-Based Approach to Mobile Robot Navigation in Unknown Environments
[attachment=32968]
Abstract
In this paper, a neurofuzzy-based approach is pro- posed, which coordinates the sensor information and robot motion together. A fuzzy logic system is designed with two basic behaviors, target seeking and obstacle avoidance. A learning algorithm based on neural network techniques is developed to tune the parameters of membership functions, which smooths the trajectory generated by the fuzzy logic system. Another learning algorithm is developed to suppress redundant rules in the designed rule base. A state mem- ory strategy is proposed for resolving the “dead cycle” problem. Under the control of the proposed model, a mobile robot can ade- quately sense the environment around, autonomously avoid static and moving obstacles, and generate reasonable trajectories toward the target in various situations without suffering from the “dead cycle” problems. The effectiveness and efficiency
INTRODUCTION
OBILE robot navigation is an essential issue in robotics and artificial intelligence. Real-time navigation is an
easy task for human beings or animals, but very difficult for robots, especially in unknown and changing environments. For real-time autonomous navigation, the robot should be capable of sensing its environment, interpreting the sensed information to obtain the knowledge of its location and the environment, planning a real-time trajectory from an initial position to a tar- get with obstacle avoidance, and controlling the robot direction and velocity to reach the target.
Neural network-based approaches have been employed for robot motion planning. In these approaches, the robot is treated as a point moving under the influence of an artificial neural potential field. The attractive potential force attracts the robot toward the target configuration, while repulsive potential forces push it away from obstacles. Dahm et al. [2], [4] used a neu- ral field approach described by an integrodifferential equation, which can be discretized to obtain a nonlinear competitive dy- namical system affecting a set of artificial neurons.
PROPOSED APPROACH
In this section, the overall control structure of the proposed neurofuzzy model is first described. The fuzzy controller is then designed. After that, the neural network-based learning algorithms are developed to tune the model parameters, and to suppress redundant rules. Finally, the state memorizing strategy to resolve the “dead cycle” problem is presented.
Overall Control Structure
To control a mobile robot to reach its destination with obsta- cle avoidance, sensors must be mounted on the robot to sense the environment and interpret the sensed information. The main sensors of the mobile robot are shown in Fig. 1. The robot is employed to test the proposed fuzzy logic-based system. The robot has two front coaxle wheels driven by different motors separately, and a third passive omnidirectional caster.
SIMULATION STUDIES
To demonstrate the effectiveness of the proposed fuzzy logic- based controller, simulations using a mobile robot simulator (MobotSim Version 1.0.03 by Gonzalo Rodriquez Mir) are per- formed. The robot is designed as shown in Fig. 1. The diameter of the robot plate is set to 0.25 m, distance between wheels is set as 0.18 m, wheel diameter is 0.07 m, and wheel width is
0.02 m. In addition to the target sensor and the speed odometer, there are nine ultrasonic sensors mounted on the front part of the robot. The angle between sensors is 20◦ . The sensor ring radius is 0.1 m. The radiation cone of the sensors is 25◦ . The sensing range of the ultrasonic sensors is from 0.04 to 2.55 m. The upper bound of the wheel speed is 0.15 m/s. In every case, the environment is assumed to be completely unknown for the robot, except the target location; and the sensing range of the onboard robot sensors are limited.
CONCLUSION
In this paper, a novel fuzzy logic-based control system, com- bining sensing and a state memory strategy, is proposed for real- time reactive navigation of a mobile robot. Under the control of the proposed fuzzy logic-based model, the mobile robot can autonomously reach the target along a smooth trajectory with obstacle avoidance. Several features of the proposed approach are summarized as follows.
1) The proposed model keeps the physical meanings of the variables and parameters during the processing, while con- ventional models [10] cannot.
2) Forty-eight fuzzy rules and two behaviors are designed in the proposed model, much fewer than conventional ap- proaches that use hundreds of rules (e.g., 600 rules in [5]).
3) The structure of the proposed fuzzy logic-based model is very simple with only 11 nodes in the first layer of the structure, while hundreds of nodes are necessary in some conventional fuzzy logic-based models (e.g., 240 nodes needed in the first layer in [10]).
4) The proposed selection algorithm can automatically sup- press redundant fuzzy rules when there is a need.
5) A very simple yet effective state memory strategy is de- signed. It can resolve the “dead cycle” problem existing in some previous approaches [6] without changing the fuzzy control rule base.
[attachment=32968]
Abstract
In this paper, a neurofuzzy-based approach is pro- posed, which coordinates the sensor information and robot motion together. A fuzzy logic system is designed with two basic behaviors, target seeking and obstacle avoidance. A learning algorithm based on neural network techniques is developed to tune the parameters of membership functions, which smooths the trajectory generated by the fuzzy logic system. Another learning algorithm is developed to suppress redundant rules in the designed rule base. A state mem- ory strategy is proposed for resolving the “dead cycle” problem. Under the control of the proposed model, a mobile robot can ade- quately sense the environment around, autonomously avoid static and moving obstacles, and generate reasonable trajectories toward the target in various situations without suffering from the “dead cycle” problems. The effectiveness and efficiency
INTRODUCTION
OBILE robot navigation is an essential issue in robotics and artificial intelligence. Real-time navigation is an
easy task for human beings or animals, but very difficult for robots, especially in unknown and changing environments. For real-time autonomous navigation, the robot should be capable of sensing its environment, interpreting the sensed information to obtain the knowledge of its location and the environment, planning a real-time trajectory from an initial position to a tar- get with obstacle avoidance, and controlling the robot direction and velocity to reach the target.
Neural network-based approaches have been employed for robot motion planning. In these approaches, the robot is treated as a point moving under the influence of an artificial neural potential field. The attractive potential force attracts the robot toward the target configuration, while repulsive potential forces push it away from obstacles. Dahm et al. [2], [4] used a neu- ral field approach described by an integrodifferential equation, which can be discretized to obtain a nonlinear competitive dy- namical system affecting a set of artificial neurons.
PROPOSED APPROACH
In this section, the overall control structure of the proposed neurofuzzy model is first described. The fuzzy controller is then designed. After that, the neural network-based learning algorithms are developed to tune the model parameters, and to suppress redundant rules. Finally, the state memorizing strategy to resolve the “dead cycle” problem is presented.
Overall Control Structure
To control a mobile robot to reach its destination with obsta- cle avoidance, sensors must be mounted on the robot to sense the environment and interpret the sensed information. The main sensors of the mobile robot are shown in Fig. 1. The robot is employed to test the proposed fuzzy logic-based system. The robot has two front coaxle wheels driven by different motors separately, and a third passive omnidirectional caster.
SIMULATION STUDIES
To demonstrate the effectiveness of the proposed fuzzy logic- based controller, simulations using a mobile robot simulator (MobotSim Version 1.0.03 by Gonzalo Rodriquez Mir) are per- formed. The robot is designed as shown in Fig. 1. The diameter of the robot plate is set to 0.25 m, distance between wheels is set as 0.18 m, wheel diameter is 0.07 m, and wheel width is
0.02 m. In addition to the target sensor and the speed odometer, there are nine ultrasonic sensors mounted on the front part of the robot. The angle between sensors is 20◦ . The sensor ring radius is 0.1 m. The radiation cone of the sensors is 25◦ . The sensing range of the ultrasonic sensors is from 0.04 to 2.55 m. The upper bound of the wheel speed is 0.15 m/s. In every case, the environment is assumed to be completely unknown for the robot, except the target location; and the sensing range of the onboard robot sensors are limited.
CONCLUSION
In this paper, a novel fuzzy logic-based control system, com- bining sensing and a state memory strategy, is proposed for real- time reactive navigation of a mobile robot. Under the control of the proposed fuzzy logic-based model, the mobile robot can autonomously reach the target along a smooth trajectory with obstacle avoidance. Several features of the proposed approach are summarized as follows.
1) The proposed model keeps the physical meanings of the variables and parameters during the processing, while con- ventional models [10] cannot.
2) Forty-eight fuzzy rules and two behaviors are designed in the proposed model, much fewer than conventional ap- proaches that use hundreds of rules (e.g., 600 rules in [5]).
3) The structure of the proposed fuzzy logic-based model is very simple with only 11 nodes in the first layer of the structure, while hundreds of nodes are necessary in some conventional fuzzy logic-based models (e.g., 240 nodes needed in the first layer in [10]).
4) The proposed selection algorithm can automatically sup- press redundant fuzzy rules when there is a need.
5) A very simple yet effective state memory strategy is de- signed. It can resolve the “dead cycle” problem existing in some previous approaches [6] without changing the fuzzy control rule base.