Localizability Estimation For Mobile Robot And Its Applications

Localization for mobile robots is one of the most important technology in robotics. A reliable self-localization performance is the basis for the robot to complete different tasks. However, in complex crowded environments, the observations may be interfered because of the map structure, unknown obstacles and other factors, which lead to the decrease of localization accuracy and sometimes even results in the failure of localization.How to evaluate the localizability of robot in a crowded environment and prevent the failure of localization is an important issue in localization research. This paper focuses on the above issue, make an in-depth study on the localizability evaluate and self-localization in crowded environment.A localizability estimation method is proposed for the widely used probabilistic grid map. Firstly, the Fisher information matrix (FIM) of robot localization is transformed to discrete form, and a static localizability matrix is obtained which is applicable for off-line estimation based on known grid map. On this basis, considering impact factor of local sensed obstacles to the static localizability matrix, a dynamic localizability matrix (DLM) is proposed for on-line estimation to deal with unexpected dynamic changes of environments. This matrix describes both the localizability index and localizability direction of mobile robot quantitatively. The results of real robot experiments under different typical environments demonstrate the validity of the proposed method.Based on the DLM, an amended-particle-filter localization algorithm is proposed to deal with the noise interference of observations in crowed environment. This algorithm estimates the DLM of the observation model and the covariance matrix of prediction model as weight indicators. Comparing these indicators, the value based on observation model to amend the predictive position is adjusted dynamically to get the final robot position. With a low DLM of the observation model, the algorithm follows the prediction model more closely to deal with unexpected changes of robot pos, ensuring the stability of localization. Otherwise, this algorithm follows the observations more closely to eliminate the accumulated error.In this paper, the intelligent wheelchair made by ourselves named Jiao Long is applied for experiment parts. Experiments under different typical environments are designed for localizability estimation, localization and navigation. The Comparative experiments demonstrate the validity of the DLM and the reliability of amended-particle-filter localization algorithm to prevent localization failure in crowed environment, which establishes stable basis for the practicality of intelligent wheelchair.