A Vision-based Monte Carlo Self-localization System on a Walking Robot

This paper presents a vision-based localization system on legged robots.Based on Monte Carlo localization,the system makes use of high-noised sensor data of artificial and natural landmarks,along with odometry data without feedback.Special methods to deal with similar natural landmarks and to incorporate information from several landmarks,as well as a new method for weight calculation and particle resampling are presented.Results of experiments on real robots show that the system can adapt to a dynamically uncertain environment,remain accurate and stable under a high noised condition,solve the problem of kidnapped robots,and perform in real-time.