Localization for Legged Robot with Single Low-Resolution Camera using Genetic Algorithm

Localization is an important area in autonomous robotics. Recently, many researchers focus on legged robot due to its ability to navigate on undulating terrain. Localization of legged robot is more difficult than that of wheeled robot in the following senses: 1) leg slippages are common during walking motion, which degrade the motion accuracy and 2) due to the oscillated walking motion of robot, sensor data are fluctuating that the degree of freedom (DOF) of legged robot is higher than that of wheeled robot. Among different kinds of sensors, single camera equipped on a legged robot is considered in this paper as camera is a low-cost and small sensor that is suitable for small-sized robot. The challenges of vision-based localization for legged robot include real-time processing, high dimensional movement, limited field of view and the lack of depth information in images. In this paper, a robust and flexible vision-based localization algorithm for legged robot using genetic algorithm (GA) is presented. The localization problem is modeled as a high dimensional optimization problem. Given a set of feature coordinates and their corresponding image points extracted from an image, position and orientation of the robot can be estimated from a single image by the proposed localization algorithm.