Figure 1. The three calibration axes of a robot
Contractors:
CSEM (CH),
KUL (B),
Krypton (B)
Funding:
Basic Technology Research Programme
Predictable, safe and cost efficient operation of a robotic device for space applications can best be achieved by programming it off-line during the preparations for the mission. Computer aided design techniques are used to assure that the movements of the robot are predictable. A software model of the robot and its work cell is made and this must be compatible with the model of the environment in which the robot must perform. Cost efficiency requirements dictate that a robot be calibrated, after which its performance must be checked against specified requirements.
Proper use of miniaturised sensing technology is needed to produce a robot of minimum size, power consumption and mass. This often requires minimising the number and types of sensors needed, and maximising the information (such as position, velocity, acceleration) which is gained from each sensor.
To study methods of assessing the performance of a robot, choosing its sensors and performing calibration and test, ESA passed a contract with industry. Results of this work are applicable to any robot whose kinematic chain needs accurate geometrical modelling.
The objectives of the robot performance assessment are:
The performance of the robot is assessed by making mathematical models of the characteristics of the error sources in each of its sub systems such as the joint, the robot link or its gripper. From these the effects of errors on the positioning accuracy of the end-effector (the functioning tip of the robot arm) can be evaluated.
Error sources are identified by a bottom-up analysis, which takes account of the capabilities of state-of-the-art production technology. For each robot sub-system, the error sources are identified and are sorted into three categories:
Once the error sources have been classified and their magnitudes defined, various statistical methods may be used to evaluate their effects when they work in combination. Simply adding the all errors together, takes no account of their statistical nature, and gives an estimate which is safe but unduly pessimistic; misapplication of statistics can produce an estimate which is too optimistic.
The accuracy of some pointing mechanisms is frequently estimated by separately estimating the root-mean-square value of each of the three error types identified above and adding them. In the case of the AMTS project, all error sources were considered as statistical variables and a single root-mean-squared error at the end-effect or was of interest.
The bottom-up approach used to establish the contribution of each error source was validated taking the case of the manipulator shown in Figure 1 as an example. for which a worst case accuracy of 2.7 mm was predicted. This was very close to its average accuracy of 2 mm.
Figure 1. The three calibration axes of a robot
Figure 2. Robot test setup (courtesy of Krypton)
If the performance prediction has shown that calibration is needed to compensate for errors, a proper calibration approach is required.
Ideally, all calibration must be done on-ground; in-orbit calibration procedures should be limited to cross-checking the validity of the model developed on-ground and, if necessary, correcting for effects such as micro-slip page or pressure gradient.
To keep the flight hardware simple, the in-orbit calibration should be achieved using only the sensors already available in the robot.
Calibration is performed in five steps:
A method for calibrating each axis independently has been successfully developed in the frame of the contract. This method uses independent measurements of motion along each of the three axes (Figure 1). The advantage of this approach compared to others, such as those requiring all robot joints move simultaneously, is that it sub-divides the general problem of robot calibration into a set of problems of lower complexity, thus achieving good stability and numerical precision. The calibration software is parametric and is suitable for calibrating any open robot kinematics chain.
As part a verification procedure, specific performance tests were carried out on a robot by Krypton under contract to ESA. The first was an accuracy test in which the robot had to adopt aspecified pose and aim at a point, a key characteristic for predictable off-line programming. The second test evaluated the repeatability with which the robot could reach a pose it had been taught to adopt; this is essential for performing repetitive and routine tasks.
Finally the multi-directional pose accuracy was tested to establish the effects of random errors and to establish the limits of the calibration procedure. The performance of the robot was measured before and after calibration. Figure 2 shows the measurement system setup, which employed three linear video cameras.
Procedures for calibrating robots both on-ground and in-orbit have been developed, and the performance of robotic devices has been successfully tested. The robot calibration procedure proved to work well, resulting in an improvement in performance by a factor of ten in some cases. The calibration software is versatile and it can beused to calibrate and evaluate most kinematic chains ranging from a simple two-axis antenna gimbal mechanism to a 10-axis manipulator. These software procedures are now used by Krypton for applications in the motor industry and else where.
Preparing for the Future Vol. 4 No. 4.