Robot Tracking using the Particle Filter and SOM in Networked Robotic Space

Detection of moving objects has been utilized in industrial robotic systems, for example, in the recognition and monitoring of unmanned systems that also require compression of moving images (Senior, 2002, Bierlaire & Antonini, 2003). Trajectory prediction of moving objects is required for a mobile manipulator that aims at the control and observation of motion information such as object position, velocity, and acceleration. Prediction and estimation algorithms have generally been required for industrial robots. For a simple example, in a pick-and-place operation with a manipulator, the precise motion estimation of the object on the conveyor belt is a critical factor in stable grasping (Nummiaro, et.al, 2002, Allen, et.al, 1992). A well-structured environment, such as the moving-jig that carries the object on the conveyor belt and stops when the manipulator grasps the object, might obviate the motion estimation requirement. However, a well-structured environment limits the flexibility of the production system, requires skillful designers for the jig, and incurs a high maintenance expense; eventually it will disappear from automated production lines. To overcome these problems, to tracking a moving object stably without stopping the motion, the trajectory prediction of the moving object on the conveyor belt is necessary (Ma, Kosecka & Sastry, 1999). The manipulator control system needs to estimate the most accurate position, velocity, and acceleration at any instance to capture the moving object safely without collision and to pick up the object stably without slippage. When the motion trajectory is not highly random and continuous, it can be modeled analytically to predict the near-future values based on previously measured data (Choo & Fleet, 2001). However, this kind of approach requires significant computational time for high-degrees-of-freedom motion, and its computational complexity increases rapidly when there are many modeling errors. In addition, performance is highly sensitive to the change of the environment. Those state-of-the-art techniques perform efficiently to trace the movement of one or two moving objects but the operational efficiency decreases dramatically when tracking the movement of many moving objects because systems implementing multiple hypotheses and multiple targets suffer from a combinatorial explosion, rendering those approaches computationally very expensive for real-time object tracking (Anderson & Moore, 1979).