A geometric approach for kinematic identification of an industrial robot using a monocular camera

•Singular Value Decomposition based method for estimating the joint axis vector for prismatic and revolute prismatic joints. This was helpful in identifying complete set of DH parameters, i.e., the set of joint variables θ's as well (corresponding to the home position about which the end-effector was moved by actuating each joint).•A method to index the measurement errors and subsequently predict the accuracy of estimated kinematic parameters.•To overcome the limited field of view of a camera, marker map based approach was adopted for the first time in experiment for kinematic identification.•The success of proposed method in the identification of kinematic parameters using monocular vision is illustrated by comparing the results with those obtained using the laser tracker and total station.•One of the outcome of this paper is the measurement of the six-dimensional pose using a monocular camera mounted on the robot that was utilized to calculate the kinematic parameters. This can be used in finding the repeatability of a mechanical systems that has poor repeatability than that of the camera used. We propose a generic formulation to identify the kinematic parameters of an industrial robot using a geometric approach when no prior information about the robot’s kinematics is available. The joint axes were estimated using the singular value decomposition applied to the pose data of the robot’s end-effector. These data were obtained by actuating one joint of the robot at a time. The approach is first illustrated using the CAD model of the robot. Next, a simulation study was performed to decide on the number of data points and angular actuation required by a revolute joint to estimate the kinematic parameters. This was done considering the fact that there exist noise in the measurements from the sensors. In this paper, we used a monocular camera mounted on the end-effector of an industrial robot KUKA KR5 Arc as the measurement sensor. We used ArUco Marker maps instead of a single calibration grid to enhance the range of actuation for each joint that was otherwise restricted due to the Field of View of the camera. The robot’s kinematic parameters were then identified using the proposed approach. The identified parameters using the monocular camera were compared with those obtained using other measurement devices, namely, a total station and a laser tracker.