A SENSOR USED FOR MEASUREMENTS PRODUCTION ROBOTS

Transcription

1 A SENSOR USED FOR MEASUREMENTS PRODUCTION ROBOTS IN THE CALIBRATION OF LOUIS J. gveren AND THOMAS I. IVES Department of Mechanical Engineering, Texas A&M University, College Station, Texas ABSTRACT A sensor has been developed and successfully implemented to automate robot calibration. Its simple design makes it ideal for production environments. It works in conjunction with precision spheres that are mounted in the workspace. The positions of the spheres are measured in advance, and the sensor is positioned over them automatically and precisely. The positioning of the sensor is gavemed by three light beams that define an exact position with respect to the sensor. This eliminates the problems experienced by other short range devices in that it acts digitally and without contact. Design of the sensor and its autonomous control from a PC are presented along with proven measurement tactics and results of performance tests. I INTRODUCTION used are as follows: laser interferometry [l]; coordinate measuring machines; theodolites [2]; time of flight devices [3]; camera devices; and short range devices like micrometers, and linear voltage digerential transformers, i.e. LVDTs. Other methods, which muld be classified as constrained methods, work by moving the robot along known constrained motions such as a predetermined surface while its servo motors are de energized. One good example of these methods would be Driels' ball bar method [4]. In this method, a ball and socket joint with a bar attached to and extending from the ball are attached to the robot's end manipulator. At the other end of the bar is another ball and socket joint whose socket remains stationary. While this is a good device in theory, one should ask how well it fits into a production environment. New hardware must be installed on site next to the robot. This is not extremely difficult or time consuming, but the calibration process using this or similar devices require the robot to make contact with the device. The design goal herein is to obtain robot position measurements automatically without contacting external fixtures. Also, maintaining an expensive measurement device, like some of those listed above, next to a production robot for a calibration that is performed at most once every few months is very uneconomical. Moving such a device over to the robot's production area as needed may be impractical. Therefore, a small, simple and inexpensive measurement device, (SSIMD), that will not require alterations of the production setup is desirable. The short range devices (i.e. micrometers, etc.) listed above work in coordination with furtures that have been placed in the workspace. The positions of these iixhms are measured with respect to the workspace. A short range device, or SSIMD, can be treated as a tool and can be stored along with other tools in a robot tool-rack. A pre measured array of fixtures, PMFA, can be arranged on a mobile workspace such that an automatic storage While the field of robot calibration continues to progress, measuring the pose, (position and orientation), and retrieval system, ASS, of a production setup brings of a robot's tool point in a production environment will the PMFA to the robot when necessary. When become increasingly important. Some devices currently calibration is complete, the PMFA is stored for later use /93 $ IEEE 74

2 However, there are setbacks to the existing measurement devices being used. Vision systems have inadequate resolution. Time of flight devices such as sonar require complex and expensive electronics, and accuracy is susceptible to ambient condition changes and echoes. Contact devices such as LVDT's and micrometers require excessive measurements and are susceptible to errors such as contact force. Therefore, a noncontact, digital sensor that can be automatically controlled with a simple control scheme and that operates with good repeatability is desirable. SENSOR CONSTRUCTION A sensor owning the above characteristics has been designed by Everett and Hsu [5]. It has been used for five years in the Texas A&M University's Mechanical Engineering Robotics Lab and has yielded good results when implemented in robot calibration. The sensor shown in Figure utilizes three LED/photo transistor pairs as optical switches. Each switch utilizes an LED that shines through an optical fiber. The light leaves the fiber and shines across a gap. The beam is received on the other side by another fiber. The light travels through this fiber and is received by a photo transistor. When enough light is blocked, the transistor becomes unbiased and turns off. do not move during the measurements. The reason for this will be understood through the discussion below. To understand the measurement process, imagine the three light beams to be solid rods with gravity always acting in the x direction of the sensor (see figure ). Also, consider a.5 inch diameter sphere. It is obvious that this sphere could remain stable, resting on the rods. The ball's center would return to the exact location every time if it was repeatedly lifted and set back down. Now consider the light beams. Imagine the sensor gap advancing slowly toward a sphere as shown in Figure 2. There exists one line of approach that will break all of the beams at the same time, and there exists one point on this line where the beams are barely broken. This will be called the trip point of the sensor. The trip point is defined as the origin of a coordinate system fixed in the sensor, and it is constant with respect to the tool of the robot. Now consider a set of precision spheres that are all the same size; the positions of their centers are known. If the sensor is tripped by a sphere, the center of that sphere is regarded as the trip point of the sensor. The trip point will be the same for all of the other precision spheres. =/ - SIDE VIEW Figure : The Sensor. TOP VIEW The sensor can be easily machined, and the fiber optics and LED/photo transistor units, LEDPT's, are readily available at reasonable prices. The fiber optics are mounted such that the beams shine across the gap and are received at the opposite side as shown in Figure. Beams and 2 are approximately 38 inches apart and roughly parallel to one another. Beam 3 is mounted approximately 3 degrees away from them. The angles of the beams and their alignment are not critical. What is important is that the fibers be mounted securely so they Figure 2: Sensor Approaching a Precision Sphere. The sensor, with its optical fibers and LEDPT's, is fabricated to be carried by the robot as a tool. Because the sensor outputs are buffered ltl logic signals, there is enough drive capability and signal to noise ratio for the outputs to be connected through a quick change mechanism if necessary. Precision spheres are mounted to a mobile fixture as shown in Figure 3. This fixture might be carried to the robot workspace by an ASRS. This mobile fixture will be referred to as a tombstone for the remainder of the paper. The spheres are mounted on the tombstone, and the positions of their centers (relative to each other) are determined. 75

3 I Figure 3: Precision Spheres Mounted on a Tombstone. SENSOR - SYSTEM INTEGRATION All the connections for sensor movement control are illustrated in Figure 4. Outputs from the photo transistors are wired into a personal computer's (PC's) digital input. It is possible to connect the sensor directly to the robot controller, and this has been done on an IJ3M To go directly into the robot controller, the controller must have three digital inputs. However, the PC was chosen for this work to achieve a higher degree of hardware independence. In the latest implementation of the sensor, the primary control program was written in C and was run on a PC. This C control program can communicate with any robot controller language that can read from a serial port. In our latest study, the program communicated with a GMF robot controller over a RS232 serial communication line. Single characters sent to the GMF robot controller represent an action to be performed by the robot. These actions are listed in Table. Some of the characters sent to the GMF controller represent two orthogonal moves, because certain beam combinations warrant both of these moves, and the trip point search time is significantly reduced by including them in one step. The robot controller's program informs the control program on the PC when the action is complete. It is best to incorporate a variable robot resolution when the sensor is searching for the trip point. This allows the robot to move quickly at the beginning of its search, then, when the sensor is lined up better, the step size can be decreased until the final movements use the finest resolution available on the robot. I Figure 4: Control Scheme for Automatic Sensor Movement. Character sent to GMF X Y Z j k V W Action Performed by GMF Controller Move one step in +x wrt robot's tool Me. Move one step in +y and +z wrt Move one step in +z and -x wrt Move one step in -x wrt robot's tool frame. Move one step in -y and +z wrt Move one step in -z and -x wrt Move one step in +y wrt robot's tool frame. Move one step in -y wrt robot's I tool frame. a I Adjust step size to one third of the r current size. Send joint angle data, set step to largest step, and back off 5". SENSOR OPERATION When the robot needs to be calibrated, the tombstone is positioned in the workspace, and the sensor is either picked up by the robot from a tool-rack or put in the end manipulator by an operator. The operator either moves the sensor near a sphere, or the robot moves to a previously taught point near the sphere. Due to safety concerns in our work, the robot is moved near each sphere manually. Once the sensor is positioned within an acceptable location, the automatic control is started and the control scheme proceeds as follows: ) The control program looks at the sensor input to determine which photo transistors are broken. 76

4 2) The information from ) is used to determine which direction the robot should move the sensor. 3) The move direction from 2) is communicated to the robot's controller and the robot moves a small amount. 4) Step one is repeated. There are a couple of limitations that can prevent finding the exact trip point. The first limitation is sensor orientation. If the orientation of the sensor frame is not defined well in the robot controller, the robot will not move along the sensor's axes with sufficient precision to simultaneously break the beams during the finest resolution moves of the robot. The second limitation is robot resolution. Even if the orientation of the sensor is well defined, the resolution of the robot may prevent this line of approach from being found exactly. The controlling code is written with these limitations in mind, and an algorithm was developed that searches until an acceptably precise trip point is found. It is also important to incorporate a check in the control program to see if the sensor movement is caught in a continuous loop. If it is, a resolution change, or a small move in another direction, has never failed to break it out of such a cycle. SENSOR PERFORMANCE In the first study performed with the sensor, ambient lighting was changed to test the effect it had on the trip point location. First, one LED beam was blocked as little as possible to achieve an off state for its corresponding photo transistor. Second, another LED was blocked as much as possible without putting its corresponding photo transistor in an off state. Then, the ambient lights were modulated. There was never an instance when the lighting effected the state of the photo transistors. In the second studies performed, an IBM 7565 was used. When performing accuracy studies with the sensor, it is important to note that its accuracy cannot be better than the finest resolution of the robot or driving system being used. Therefore, some of the error in these tests is due to the repeatability of the robot. The sensor was held by the 7565 and repetitively tripped. Each time, the manipulator wrist was held constant, and the translation positions were recorded. A worst case repeatability error of.32 inches resulted when the LEDPT's were set to their lowest sensitivity settings. When the LEDPT's were at their most sensitive settings, the repeatability error was.23 inches. To further evaluate the sensorhobot performance as a system, the sensor was installed on a GMF S- R. We were specifically interested in determining the sensor's ability to repeatibly find a unique trip point for a constant sensor Orientation. To independently measure the performance, we rigidly attached an additional set of three tooling balls to the sensor. A sketch of the sensor with the attached balls is shown in figure 5. The robot was used to trip the sensor beginning from 4 distinct locations while the sensor orientation remained approximately constant. Once tripped, the center positions of each tooling ball mounted on the sensor were measured using a coordinate measuring machine. Table 2 gives the measured positions of the three balls for each approach direction. By allowing ball to be the origin, ball 2 to be on the X axis, and ball 3 to lie in the XY plane, a coordinate system can be defined using the ball position data. Table 3 gives the poses of the coordinate system defined by the ball positions in Table 2. Each of the pose representations should be the same. If we compute the error between two poses as T,-'(T, - T,) then the errors between approaches,two and one, three and one, and four and one are as shown in Table 4. Figure 5: Sensor with Tooling Balls Attached. Approach Ball Xmm Ymm Z mm II Table 2: Tool Ball Position Measurements. 77

5 P Approach T-Matrix Table 3: Pose Matrices for Measurements in Table Error Norm Table 4: Pose Errors Calculated from Poses in Table 3. TOOL POSITION Assume that the tombstone used for calibration has three precision spheres mounted on it. The sensor is moved near the first sphere, and the automatic control of the sensor is started. Once the sensor finds the trip point, and the joint angles are recorded for this position, the sensor is moved close to this same sphere again. However, this time the sensor is in a far different orientation than it was on the previous measurement. For example, the first approach of the sensor on the sphere might have been with an 8' orientation about the world frame's z axis, the second approach with a -8' orientation about the world frame's z axis, and the last with an 8' orientation about the world frame's z axis and a 9' orientation about the world frame's x axis. This practice can be repeated with each of the spheres until the desired amount of data for the calibration has been collected. TOOL ORIENTATION It is possible to measure both orientation data and position data with this sensor, but they cannot be measured at the same time. Orientation data must be obtained by a Werent method than the position data. The robot can be calibrated using only position data. Without orientation data, the last three orientation parameters of the tool point cannot be found, but the nominal orientation of the tool point is adequate for tasks that merely require accurate positioning. Once the robot is calibrated using position data, the remaining three tool orientation parameters can be found. One method uses the fixture shown in Figure 7 which is held by the manipulator while the sensor is fixed to ground. The procedure for this method has been developed by Everett and Ong [6]. By using this fixture it is possible to determine the pose of the sensor frame to the robot end effector frame. The procedure is similar to solving the sensor registration problem [7]. Figure 7: Fixture Used to Obtain Orientation Measurements in Conjunction with the Sensor. REFERENCES [l] K. Lau, R. J. Hocken, and W. C. Haight, "Automatic laser tracking interferometer system for robot metrology," in Precision Engineering, vol. 8, pp. 3--7, January 986. [2] D. E. Whitney, C. A. Lozinski, and J. M. Rourke, "Industrial robot calibration method and results," in International Computers in Engineering Conference and Exhibit, pp. 92-,

6 [3] H. W. Stone, A. C. Sanderson, and C. P. Neuman, "Arm signature identification," in IEEE International Conference on Robotics and Automation, (San Francisco, California), pp. 4-48, April [4] M. Driels and M. Wiegand, "Passive calibration of a teleoperator mechanism," in Applications of Modeling and Identijication to Improve Machine Performance, (L. J. Everett and M. Driels, eds.), pp , ASME Winter Annual Meeting, 99, [5] T.-W. Hsu, Robot Accuracy Improvenient Through Kinematic Parameter Identification PhD Dissertation, Texas A&M University, Mechanical Engineering, May 987. [6] L. J. Everett and L. E. Ong, "Solving the generalized sensor-mount registration problem," Under Review IEEE Journal on Robotics and Automation, 992. [7] J.C.K. Chou and M. Kame], "Finding the Position and Orientation of a Sensor on a Robotics Manipulator Using Quaternions", The International Journal of Robotics Research, Volume, No. 3, pp , June