About Robotic Hockey
Dynamic Bin Packing
The solution to the challenge of bin picking in a static atmosphere has been adopted by many in industry. Dynamic bin picking is the next step. Dynamic bin picking is similar to static bin picking but the parts being picked are in motion. The dynamics of the parts being picked creates a need for the parts to be tracked. In our project the part being tracked is an air hockey puck that is moving along the surface of an air hockey table. In this case the puck will not be tracked and picked, but will be tracked and hit back to a person at the other end of the table. In essence we have created a robotic system to compete against a human in the game of air hockey, using technology that can be applied to dynamic bin picking.
The frame that supports the robot was designed in CATIA V5R17. We then fabricated the frame from steel box tubing, plate steel, and M12 bolts. The frame consists of four towers that are supported by two cross members per side, except the front where one cross member serves as support to allow room for the table. Additionally there is a truss that spans the width of the frame that the robot is mounted to. This truss is bolted to the cross members of the frame. The cross members are then bolted to the towers at the corners of the frame.
KUKA KR3 Robot
The KR3 is a six axis robot which allows for a great deal of freedom of motion for this application. The size of the KR3 makes in an optimal choice for air hockey and the size limitations of the frame and the air hockey table. The KR3 was designed for floor mounted applications as well as shelf mounted applications so the inverted position of the robot will not become a problem in its operation. The KR3 has a working payload of 3 kilograms. This means that at its maximum reach the KR3 can move a mass of 3 kilograms at its maximum linear velocity. The maximum linear velocity of the robot is 2 m/s. At 2m/s the robot is not as fast as a human so, to design a competitive system, a mechanical puck striker was needed.
The mechanical striker will strike the puck with a greater impulse force than could be achieved by the robot alone. This will increase the velocity of the puck ergo increasing the ability of the robot to score on the human. The striker was also designed in CATIA V5R17, and then printed in a rapid prototype 3-D printer. Our initial design concept for the striker is powered by an electric motor with a cam attached to the shaft. There are four spring loaded disc quarters that are struck by the cam and move outward in all directions by 6.35mm. Since the robot will not know which quadrant of the striker will be in contact with the puck, the striker is programmed to run at all times while the puck is moving toward the robot. Since the striker motor travels at around 1000 rpm, the disc quarters move out and back in about 16 times per second. In our first design there was nothing to guide the stroke of the disc quarters in and out. This allowed for too much chatter at high speed. Slots in the disc quarters and locating pins in the striker body were added to reduce chatter and decrease wear over time. The major diameter of the striker in its rest position is 100mm . This is the size of the striker that the human will be using, so not to give the robot an unfair advantage.
Air Hockey Table
The air hockey table that we are currently using is 1498.6mm (L) x 736.6mm (W). The side rails of the table are made of medium density fiberboard (MDF). The coefficient of restitution between this material and the plastic material that the puck is made from is not as high as we were hoping. Too much energy from the impact of the puck and the table was being absorbed into the table. Additionally, the MDF walls were deflecting upon impact further altering the trajectory. This was affecting the exit velocity and the deflection angle of the puck after contact with the wall. The change in velocity and direction was throwing of our algorithm just enough that it became a problem. To address this problem, aluminum rails were fabricated and attached to the sides.
The camera that we are using is a Point Grey Research Flea 2, firewire, 1/3” CCD, color camera with an operating speed of 30 frames per second. The lens on the camera has a focal length of 3.5mm. In order to select this lens for the camera, the distance from table surface to camera mount was determined to be 1220mm. A 3.5mm lens yields a field of view 10% longer than the length of the table and 83% wider than the table. The extra width is due to the camera ratio of 4/3 differing from the table ratio of closer to 7/3. The robot controller is connected to the computer that is running a VB.net application based on a VisionPro application. Communication is handled through a serial port between the robot and the computer. Near the tool mounting plate of the robot there is an I/O port that will run the striker motor. A special connector was made that allows for the motor to be turned on and off through the robot instead of being controlled by an external programmable logic controller. The computer being used to process the camera inputs and solve the algorithms is a Dell Precision 690. This machine has the capability to solve and compute information at the speed and consistency needed to operate such a complex program.of the table, increasing the efficiency of impact but decreasing the width of the playing surface by a total of 50.8mm.
For a robot to be able to play air hockey or perform dynamic bin picking, the object that is being affected by the robot must be mathematically described. The distance, velocity, direction and in some cases acceleration must be observed and calculated in a fraction of a second. In this application our object is a hockey puck that is 63.5mm in diameter. Since a hockey puck only travels in two dimensions, our equations are only concerned with the x and y directions.
Here is a PDF file of the mathematic equations used in our air hockey project.
To this point we have successfully developed the ability to defend the goal from the strike of an opponent given a reasonable velocity. The robot will adjust to the path of the puck and hit it back towards its opponent. With this progress we have finished phase one and will soon start work on phase two, offensive play. We are very close to a successful dynamic bin picking program and provided the correct resources we could develop a completely successful application. Our research and development of this technology will continue as we move into phase two. After phase two we will move on to phase three where we will develop an artificial intelligence that will learn strategy from other opponents. The direction that the robotic industry is moving suggests that there is a very bright future on the horizon in this field. A copy of the research paper we have submitted to ASEE is provided in the downloads page.