Research On Robot Localization And Obstacle Avoidance In The Navigation

This paper systematically presents the localization and obstacle avoidance navigation approaches for indoor autonomous mobile robots, and mainly on the localization approaches. The research topics include approaches of robot's pose tracking, Markov localization, Monte Carlo localization, real-time obstacle avoidance.Two different global localization methods based on Bayesian estimation theory are investigated in the paper. The first one is the Markov localization approach. it use a discrete representation of the world model where each cell contains the probability that the robot is in that cell. We compute the likelihood of every possible localization with the environment measurement data from sensors and the robot's movements measurement data from odometers. The robot's pose is first predicted by kinematics models of the robots, and then is updated by features extracted measurement model. We re-compute the likelihood of the robot being in every possible localization based on Bayesian rule given that data, therefore we use the updated pose estimation as the real robot's pose.The second one is the Monte Carlo Localization (MCL) method. MCL method is a version of Markov Localization algorithm, which is a recursive Bayesian filter that estimates the posterior distribution of robot poses conditioned on sensor data. It can represent multiple uni-modal distribution, instead of calculating probability distribution function explicitly, MCL is to represent the belief by a set of samples,which is dawn from the posterior distribution over the robot's poses. The pose estimation algorithm of MCL and the sampling method from motion model and sensors model are introduced in the paper, needs less storage space than position probability grids method, and is with higher accuracy and faster computing efficiency. The common MCL has difficulty in the kidnapped robot problem, thetime it takes for such localization may be unacceptably long. We introduce an improved approach called sensor resetting, which is robust to modeling errors including unmodelled movements and systematic errors. Sensor resetting picks a number of samples to add based on the sensor data when the most recent sensor reading does not agree with where the robot thinks it is, which means the robot is lost. It can be used in real time on systems with limited computational power. We present results from the experiments demonstrating the success of the algorithm and results from simulation comparing SRL to MCL, including a comparison with the common MCL that shows that the stated advantage of sensor resetting localization do indeed hold.This paper takes a overview of multi-robots system research and presents the design and some results of autonomous behaviors for tightly-coupled cooperation in heterogeneous robots teams, specifically for the task of assistance-localization. These cooperative behaviors enable capable, sensor-rich (leader) robot to assist in navigation of sensor-limited simple robot that have no onboard capabilities for obstacle avoidance or localization, and only minimal capabilities for kin recognition. The simple robot must be dispersed throughout a known, indoor environment to serve as a sensor network. They are unable to autonomous disperse themselves or move to planned sensor position independently. The presented cooperative behaviors enable the successful deployment of simple robots by assistance from the leader robots. Some results of the cooperative behaviors are presented.The real-time obstacle avoidance is the basis for robots to navigate reliably. A method based on multiple visual-edges information fusion is proposed for obstacle detecting and route selection. Based on the visual information of the environment, the multi-edge exacting algorithm is presented in order to obtain the Radial Obstacle Profile. According to the ROP, an advanced route selection algorithm considering the robot size and real-time velocity to navigate robot in unstructured environment is proposed. The efficiency and feasibility of the method has been proved by the physical robot experiment in different environments.