The position of a moving object can be tracked in numerous ways, the simplest of which is to use a single static sensor. However, the information from a single sensor cannot be verified and may not be reliable without performing multiple measurements of the same object. When multiple static sensors are used, each sensor need only take a single measurement which can be combined with other sensor measurements to produce a more accurate position estimate.
Work has been done to develop sensors that move with the tracked object, such as relative positioning, but this research takes this concept one step further; this dissertation presents a novel, highly capable strategy for utilizing a multi-robot network to track a moving target. The method optimizes the configuration of mobile tracking stations in order to produce the position estimate for a target object that yields the smallest estimation error, even when the sensor performance varies.
The simulations and experiments presented here verify that the optimization process works in the real world, even under changing conditions and noisy sensor data. This demonstrates a simple, robust system that can accurately follow a moving object, as illustrated by results from both simulations and physical experiments. Further, the optimization led to a 6% improvement in the target location estimate over the non-optimized worst-case scenario tested with identical sensors at the nominal fixed radius distance of 2.83 m and even more significant improvements of over 90% at larger radial distances.
This method can be applied to a wider variety of conditions than current methods since it does not require a Kalman filter and is able to find an optimal solution for the fixed radius case. To make this optimization method even more useful, it is proposed to extend the mathematical framework to n robots and extend the mathematical framework to three dimensions. It is also proposed to combine the effect of position uncertainty in the tracking system with position uncertainty of the tracking stations themselves in the analysis in order to better account for real-world conditions.
Additionally, testing should be extended to different platforms with different sensors to further explore the applicability of this optimization method. Finally, it is proposed to modify the optimization method to compensate for the dynamics of the system so that sensor systems could move into an intercept course that would result in the optimal configuration about the tracked object at the desired time step. These proposals would result in a more applicable and robust system than is currently available.
Source: Santa Clara University
Author: Jasmine Cashbaugh