An Integrated Online Path Planning and Control Approach for Robotic Sensor Networks
Repository Usage Stats
This dissertation addresses an information-driven sensor path planning problem which has various applications such as robot cleaning, environment monitoring, and manufacturing. Information-driven sensor path planning is concerned with planning the measurements of a sensor or a sensor network in order to support sensing objectives, such as target detection, classification and localization, based on prior information. When the sensor's field-of-view or visibility region is bounded, the sensor's position and orientation determine what targets can be measured at any given time. Therefore, the sensor path must be planned in concert with the measurement sequence. When sensors are installed on robotic platforms and are deployed in an obstacle-populated environment, the sensor path must also avoid collisions between the platform and the obstacles or other robotic sensors. Addressing this sensor path planning problem, this dissertation first presents a general and systematic approach for deriving information value functions that represent the expected utility of sensor decisions in a canonical sensor planning problem. The resulting information functions and search strategies are compared through extensive numerical simulations involving direct-search, alert-confirm, task-driven, and log-likelihood-ratio search strategies, and the maximum a-posteriori, maximum-likelihood,
and Neyman-Pearson decision rules. After that a novel off-line information roadmap method is developed to navigate single robotic sensor in which obstacles, targets, sensor's platform and field of view are represented as closed and bounded subsets of an Euclidean workspace. The information roadmap is sampled from a normalized information theoretic metric that favors samples with a high value of information in configuration space. Finally, when multiple robotic sensors are deployed in the workspace, and information of the workspace such as geometry, location, and prior measurements on targets and obstacles can become available online, another novel sensor path planning method, named information potential method, is proposed to take into account the new information obtained over time. Targets with high information value tend to have high probability to be measured by the robotic sensor network. A hybrid control system is utilized to coordinate and control each robotic sensor in the network to detect and measure obstacles and targets in the workspace. The potential function is also utilized to generate the milestones in a local probabilistic roadmap method to help robotic sensors escape their local minima.
The proposed methods are applied to a landmine classification problem to plan the path of a robotic sensor network in which each robot is equipped with a ground-penetrating radar. Other sensors, such as infrared sensors on unmanned aerial vehicles (UAVs), are utilized a priori for target detection and cursory classification. In the off-line sensor path planning applications for a single robotic sensor, experiments show that paths obtained from the information roadmap exhibit a classification efficiency significantly higher than that of existing robot motion strategies. Also, the information roadmap can be used to deploy non-overpass capable robots that must avoid targets as well as obstacles. Then in the multiple online robotic sensor network path planing applications, experiments show that path obtained from the information potential method takes advantages of the online information and coordination among robotic sensors, and the results show that the information potential method outperforms other strategies such as rapidly-exploring random trees and classical potential field methods that does not take target information value into account.
DepartmentMechanical Engineering and Materials Science
More InfoShow full item record
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 United States License.
Rights for Collection: Duke Dissertations