Precise self-localization of a walking robot on rough terrain using parallel tracking and mapping

Purpose - The purpose of this paper is to describe a novel application of the recently introduced concept from computer vision to self-localization of a walking robot in unstructured environments. The technique described in this paper enables a walking robot with a monocular vision system (single camera) to obtain precise estimates of its pose with regard to the six degrees of freedom. This capability is essential in search and rescue missions in collapsed buildings, polluted industrial plants, etc. Design/methodology/approach - The Parallel Tracking and Mapping (PTAM) algorithm and the Inertial Measurement Unit (IMU) are used to determine the 6-d.o.f. pose of a walking robot. Bundle-adjustment-based tracking and structure reconstruction are applied to obtain precise camera poses from the monocular vision data. The inclination of the robot's platform is determined by using IMU. The self-localization system is used together with the RRT-based motion planner, which allows to walk autonomously on rough, previously unknown terrain. The presented system operates on-line on the real hexapod robot. Efficiency and precision of the proposed solution are demonstrated by experimental data. Findings - The PTAM-based self-localization system enables the robot to walk autonomously on rough terrain. The software operates on-line and can be implemented on the robot's on-board PC. Results of the experiments show that the position error is small enough to allow robust elevation mapping using the laser scanner. In spite of the unavoidable feet slippages, the walking robot which uses PTAM for self-localization can precisely estimate its position and successfully recover from motion execution errors. Research limitations/implications - So far the presented self-localization system was tested in limited-scale indoor experiments. Experiments with more realistic outdoor scenarios are scheduled as further work. Practical implications - Precise self-localization may be one of the most important factors enabling the use of walking robots in practical USAR missions. The results of research on precise self-localization in 6-d.o.f. may be also useful for autonomous robots in other application areas: construction, agriculture, military. Originality/value - The vision-based self-localization algorithm used in the presented research is not new, but the contribution lies in its implementation/integration on a walking robot, and experimental evaluation in the demanding problem of precise self-localization in rough terrain.