Electromechanical Development of a Low Cost End Effector Pose Estimation System for Articulated Excavators

Abstract

Vision-based pose estimation, in which optical cameras monitor fiducial markers to determine the three dimensional positioning and orientation of an articulated machine's end effector, offers a promising low-cost alternative to currently available sensor packages that are non-ubiquitous and cost prohibitive for a large portion of the market. Whereas traditional sensor systems determine end effector pose via kinematic chains passing through the links of a machine, optical sensor systems are capable of determining pose by observing an end effector directly. However, since markers cannot be mounted on an excavator's bucket for occlusion and durability reasons, a short kinematic chain must be used. An electromechanical design is proposed to provide such function for a low cost marker-based excavator pose estimation system. Several iterations of design and experimentation are discussed, including a four-bar linkage system, a synchronous belt system, a bucket linkage system, and a cable potentiometer system. The four-bar linkage and toothed belt systems were designed to transmit bucket angle information to cameras through the manipulation of a marker's pose, but were found to possess Gimbal lock and practicality issues, respectively. To overcome such issues, a generalized mapping approach was adopted and implemented in a bucket linkage design and a cable potentiometer design. The viability of the cable potentiometer system was experimentally confirmed, along with the identification of further work needed to refine the technology for large-scale practical implementation.

Pose Estimation of Excavator Manipulator Based on Monocular Vision Marker System

“…However, this configuration worked with the camera’s optical axis, to be “normal” to the marker plane. Lundeen et al [ 28 , 29 ] developed an excavator pose measurement system that consists of a marker and an optical system. The system designed four solutions to estimate the final effector pose of the excavator.…”

Vision-based excavator pose estimation using synthetically generated datasets with domain randomization

“…Methods that rely only on visual data from a single or multiple cameras have also been proposed [19]. In the earlier works, fiducial markers are mounted on the equipment's points of interest [20][21][22][23][24]. This enables the ability to recognise the position and orientation of markers, thus estimating the pose of the articulated machinery.…”