The proposed navigator consists of an avoidance behavior and goal-seeking behavior. Two behaviors are independently designed at the design stage and then combined them by a behavior selector at the running stage. A behavior selector using a bistable switching function chooses a behavior at each action step so that the mobile robot can go for the goal position without colliding with obstacles. Fuzzy logic maps the input fuzzy sets representing the mobile robot's state space determined by sensor readings to the output fuzzy sets representing the mobile robot's action space. Fuzzy rule bases are built through the reinforcement learning which requires simple evaluation data rather than thousands of input-output training data. Since the fuzzy rules for each behavior are learned through a reinforcement learning method, the fuzzy rule bases can be easily constructed for more complex environments. In order to find the mobile robot's present state, ultrasonic sensors mounted at the mobile robot are used. The effectiveness of the proposed method is verified by a series of simulations. >

A sensor-based navigation for a mobile robot using fuzzy logic and reinforcement learning

This paper reviews the major contributions to the Motion Planning (MP) field throughout a 35-year period, from classic approaches to heuristic algorithms Due to the NP-Hardness of the MP problem, heuristic methods have outperformed the classic approaches and have gained wide popularity After surveying around 1400 papers in the field, the amount of existing works for each method is identified and classified Especially, the history and applications of numerous heuristic methods in MP is investigated The paper concludes with comparative tables and graphs demonstrating the frequency of each MP method’s application, and so can be used as a guideline for MP researchers Keywords—Robot motion planning, Heuristic algorithms

Classic and Heuristic Approaches in Robot Motion Planning A Chronological Review

https://ntrs.nasa.gov/api/citations/20080004624/downloads/20080004624.pdf

Method for neural network control of motion using real-time environmental feedback

An improved approach to the solution of inverse kinematics problems for robot manipulators

A recurrent neural network, called the Lagrangian network, is presented for the kinematic control of redundant robot manipulators. The optimal redundancy resolution is determined by the Lagrangian network through real-time solution to the inverse kinematics problem formulated as a quadratic optimization problem. While the signal for a desired velocity of the end-effector is fed into the inputs of the Lagrangian network, it generates the joint velocity vector of the manipulator in its outputs along with the associated Lagrange multipliers. The proposed Lagrangian network is shown to be capable of asymptotic tracking for the motion control of kinematically redundant manipulators.

A Lagrangian network for kinematic control of redundant robot manipulators

A method for robot obstacle avoidance and path planning is proposed. The algorithm is based on a camera image feedback loop utilizing a neural network for image processing. The method can successfully generate collision-free paths in a 2D robot workspace containing randomly-placed polygonal obstacles. The control algorithm is simple and robust and has low computational requirements. Controller simulation implemented on a personal computer can generate collision-free robot paths in real time, requiring approximately 1 sec. per robot move. >

Neural network approach to path planning for two dimensional robot motion

A parallel neural network architecture called a partitioned network is proposed for applications which demand high learning accuracy. The partitioned neural network is composed of a preprocessing layer and partition modules containing dedicated neurons. The learning equations used are those of the backpropagation algorithm. The network has been applied to learning of the inverse kinematic solution of a six-degree-of-freedom robot manipulator. After training, the partitioned network was able to predict robot joint angles not included in the training data set with average errors of 0.9 degrees , 3.6 degrees , 2.1 degrees , 6.9 degrees , 6.5 degrees , and 8.5 degrees for the first, second, third, fourth, fifth, and sixth joints, respectively. >

Partitioned neural network architecture for inverse kinematic calculation of a 6 DOF robot manipulator

A software-based calibration method is developed to decrease the positioning errors of a SCARA assembly robot caused by static deflection of the robot arm under force and moment load. A robot stiffness model is constructed and used to calculate differential joint angle corrections which are then applied to the inverse kinematic solution of the ideal robot model to compensate for the static deflection errors. Least squares polynomial approximation and a neural network are used for storing the joint correction values and for interpolation. A computer simulation of the calibration method demonstrated a decrease in the average positioning error from 0.43 mm to 0.03 mm and in the range of error from +or-0.35 mm to +or-0.10 mm in the calibrated area. >

C. Kozakiewicz

Papers

Neural network approach to path planning for two dimensional robot motion

Partitioned neural network architecture for inverse kinematic calculation of a 6 DOF robot manipulator

Calibration analysis of a direct drive robot