Intercepting a Falling Object: Digital Video Robot Keshav Mundhra, Anthony Suluh, Thomas Sugar, Michael McBeath Mechanical and Aerospace Engineering, Psychology Arizona State University Tempe, AZ 85283 Abstract Human based algorithms to catch fly balls have been research and studied. In this paper, the validity of the human models are tested by catching balls that are dropped vertically downward. The regular OAC (Opti- cal Acceleration Cancellation ) and the inverted OAC models are simulated first, and then tested experimen- tally with a mobile robot. The simulation and the ex- perimental results show that both methods are able to intercept the dropped ball, but the initial and final mo- tions are different. A new digital image processing pro- gram, DVRobot, was written to process digital images very quickly. 1 Introduction The digital video robot (DV Robot) is a high-speed robot programmed to catch fly balls based on human navigational principles. This high-speed system in- volves distribution of tasks between computers. Vi- sion based navigational strategies require tremendous image processing to extract desired parameters from the video. A fast computer does this task and then the information is transmitted to the computer on the robot, which uses this data in the implementation of the navigational algorithm. Digital video systems use IEEE 1394 that provides a high-speed, Plug and Play-capable bus. The goal of the protocol is to provide an easy-to-use, low-cost, high-speed communication line. The protocol is also very scalable, and provides for both asynchronous and isochronous applications. High speed video data is needed for interception tasks because the task lasts only for 1 to 5 seconds. A falling object is an interesting interception case for a mobile robot or a user because the object is seen falling during the entire task. In psychology case studies, only the interception of thrown objects which rise and fall have been studied. We are interested in modeling, simulating, and experimentally testing vi- sion based algorithms for navigating towards a falling Figure 1: An extra computer is added to the Nomad Scout for image processing. A new program, DVRobot is written which can capture and analyze images from 1394 Digital Cameras or DV Camcorders. object. 2 Literature Review The algorithm for catching dropped balls is based on the Optical Acceleration Cancellation (OAC) model [1]. In this model, the fielder selects a run- ning path to achieve optical acceleration cancellation of the ball in the image plane or constant optical rate in the camera’s image plane. For this strategy, fielders maintain their alignment with the ball and also main- tain a constant change in the tangent of the optical angle, tan(α). There are two modes to simulate the motion of the camera. In the passive model, the camera is stationary and the image of the ball in the picture should rise/fall at a constant rate if d dt (tan(α)) is to be kept constant. In active mode, the tilt of the camera is constantly