Flexible manufacturing workcells, capable of economically producing smuli lot sizes, are becoming widespread throughout industry. This, in turn, has further encour. gcd research into the development of intelligent robotic systems capable of performing a multitude of tasks autonomously, often without complete a priori information, while reacting and adapting to continuous changes in the working environment. An important research area in this field is automated motion generation for the interception of moving objects. The Active Prediction Planning and Execution (APPE) strategy provides one approach for solving this problem. Using this strategy, the motion of an object through a robot’s work-space is predicted. Robot motion to intercept the object is then planned and executed. These three stages are repeated, as necessary, to ensure successful interception.

Over the course of this research an APPE system first proposed in [Crof95] for the interception of moving objects was developed. Unlike the APPE systems reported in the literature that restrict interception to a few non-optimal interception lines, our APPE system can intercept a moving object anywhere within the robot’s work-space. A time-optimal rendezvous-point planning strategy is used to ensure that interception occurs at the earliest possible time.

For implementing the proposed APPE system, first, a vision-based target tracking and trajectory prediction module was developed. In this context, several predictive filtering techniques were reviewed, including AutoRegressive Moving Average with Exogenous input (i.e., ARMAX) models, α-β-γ filters, and Kalman filters. Because of its superior performance and its built-in confidence measure, the Kalman filter approach was adopted and, subsequently, a variety of motion models were investigated. Simulations were used to evaluate the predictive and tracking performance of each Kalman filter and to select a filter well suited to our experimental test-bed.

The second important module within the APPE system, namely the robot-motion controller, required the implementation of the high-level time-optimal motion-planning algorithms developed in [Crof95] on a PC-based real-time system. Furthermore, as a necessary modification, an additional computationally-efficient, quintic-polynomial-based Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. robot-motion-planning algorithm was developed- This algorithm allows for trajectory-replanning rates compatible with video-frame rates achievable by the prediction system.

An APPE moving-object interception experimental system, consisting of a computer vision system, a six degree-of-freedom industrial robot, and a pair of personai computers, was built, and experiments were successfully conducted. To assess further the applicability of this system, a tracking-based moving-object-interception system was also developed. Rather than predicting the target’s future path and then planning robot motion to an optimal rendezvous point as the APPE system does, the tracking-based system simply plans robot motion to the target’s predicted current position. Simulations and experimental trials showed that for non-predictable target trajectories, the performances of both techniques are very similar, since both systems are capable o f rapid replanning. However, for predictable target motions, such as constant-velocity paths, the APPE object-interceptor was shown to be able to intercept the target sooner.