No description available.

No description available.

No description available.

<p>Tactile sensing is an essential human ability for understanding their surroundings. It allows humans to detect and manipulate objects that are concealed or difficult to see in low-light settings. Further, tactile sensing enables people to comprehend object and surface properties that cannot be obtained through visual feedback alone. This is achieved with gentle touches, enabling tactile exploration of fragile, sensitive objects, or living organisms. This capability could be transferred to robots through suitable hardware and algorithms. Nevertheless, current tactile sensors and skills for robotics are not comparable to the tactile sense of humans, thus resulting in inferior characterization of scenes and a risk of altering object states.</p> <p><br></p> <p>To address these limitations, this dissertation proposes a novel framework for robot active tactile exploration and object characterization. The framework combines bioinspired soft sensors and minimally invasive tactile exploration strategies to minimize perturbations to objects. This framework was achieved by: (1) an ultrasensitive whisker sensor that enables object characterization with minimal interaction forces; (2) autonomous tactile exploration skills to localize objects and then characterize their shape and surface properties; and (3) machine learning techniques to analyze contact information gathered by our tactile sensors, enabling the understanding of object attributes by tactile sensing alone. </p> <p><br></p> <p>Experiments were conducted to validate the effectiveness of the framework. In terms of object localization efficiency, informative path planners and contour exploration patterns outperformed baseline methods. Furthermore, the whisker sensor was successfully employed to characterize object surface and liquid properties. Finally, the features found through the characterization process allowed for successful classification by machine learning techniques. These results indicate that the proposed framework can effectively gather multimodal features from environments while maintaining the safety of objects. </p>

京都大学 / 新制・課程博士 / 博士(工学) / 甲第24604号 / 工博第5110号 / 新制||工||1978(附属図書館) / 京都大学大学院工学研究科機械理工学専攻 / (主査)教授 松野 文俊, 教授 小森 雅晴, 教授 神田 崇行 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DGAM

This thesis addresses the improvement of cooperative task allocation to vehicles in multiple-vehicle, multiple-target scenarios through the use of more effective preplanned tours. Effective allocation of vehicles to targets requires knowledge of both the team objectives and the contributions that individual vehicles can make toward accomplishing team goals. This is primarily an issue of individual vehicle path planning --- determining the path the vehicles will follow to accomplish individual and team goals. Conventional methods plan optimal point-to-point path segments that often result in lengthy and suboptimal tours because the trajectory neither considers future tasks nor the overall path. However, cooperation between agents is improved when the team selects vehicle assignments based on the ability to complete immediate and subsequent tasks. This research demonstrates that planning more efficient tour paths through multiple targets results in better use of individual vehicle resources, faster completion of team objectives, and improved overall cooperation between agents. This research presents a method of assigning unmanned aerial vehicles to targets to improve cooperation. A tour path planning method was developed to overcome shortcomings of traditional point-to-point path planners, and is extended to the specific tour-planning needs of this problem. The planner utilizes an on-line learning heuristic search to find paths that accomplish team goals in the shortest flight time. The learning search planner uses the entire sensor footprint, more efficiently plans tours through closely packed targets, and learns the best order for completion of the multiple tasks. The improved planner results in assignment completion times that range on average between 1.67 and 2.41 times faster, depending on target spread. Assignments created from preplanned tours make better use of vehicle resources and improve team cooperation. Path planning and assignment selection are accomplished in near real-time through the use of path heuristics and assignment cost estimates to reduce the problem size to tractable levels. Assignments are ordered according to estimated or predicted value. A reduced number of ordered assignments is considered and evaluated to control problem growth. The estimates adequately represent the actual assignment value, effectively reduce problem size, and produce near-optimal paths and assignments for near-real-time applications.

Unmanned air vehicles (UAVs) have become an important area of research. UAVs are used in many environments which may have previously unknown obstacles or sources of danger. This research addresses the problem of obstacle mapping and path planning while the UAV is in flight. Online obstacle mapping is achieved through the use of a multi-resolution map. As sensor information is received, a quadtree is built up to hold the information based upon the uncertainty associated with the measurement. Once a quadtree map of obstacles is built up, we desire online path re-planning to occur as quickly as possible. We introduce the idea of a quadtree rapidly-exploring random tree (RRT), which will be used as the online path re-planning algorithm. This approach implements a variable sized step instead of the fixed-step size usually used in the RRT algorithm. This variable step uses the structure of the quadtree to determine the step size. The step size grows larger or smaller based upon the size of the area represented by the quadtree it passes through. Finally this approach is tested in a simulation environment. The results show that the quadtree RRT requires fewer steps on average than a standard RRT to find a path through an area. It also has a smaller variance in the number of steps taken by the path planning algorithm in comparison to the standard RRT.

Unmanned aircraft systems (UAS) represent the future of modern aviation. Over the past 10 years their use abroad by the military has become commonplace for surveillance and combat. Unfortunately, their use at home has been far more restrictive. Due to safety and regulatory concerns, UAS are prohibited from flying in the National Airspace System without special authorization from the FAA. One main reason for this is the lack of an on-board pilot to "see and avoid" other air traffic and thereby maintain the safety of the skies. Development of a comparable capability, known as "Sense and Avoid" (SAA), has therefore become a major area of focus. This research focuses on the SAA problem as it applies specifically to small UAS. Given the size, weight, and power constraints on these aircraft, current approaches fail to provide a viable option. To aid in the development of a SAA system for small UAS, various simulation and hardware tools are discussed. The modifications to the MAGICC Lab's simulation environment to provide support for multiple agents is outlined. The use of C-MEX s-Functions to improve simulation performance and code portability is also presented. For hardware tests, two RC airframes were constructed and retrofitted with autopilots to allow autonomous flight. The development of a program to interface with the ground control software and run the collision avoidance algorithms is discussed as well. Intruder sensing is accomplished using a low-power, low-resolution radar for detection and an Extended Kalman Filter (EKF) for tracking. The radar provides good measurements for range and closing speed, but bearing measurements are poor due to the low-resolution. A novel method for improving the bearing approximation using the raw radar returns is developed and tested. A four-state EKF used to track the intruder's position and trajectory is derived and used to provide estimates to the collision avoidance planner. Simulation results and results from flight tests using a simulated radar are both presented. To effectively plan collision avoidance paths a tree-branching path planner is developed. Techniques for predicting the intruder position and creating safe, collision-free paths using the estimates provided by the EKF are presented. A method for calculating the cost of flying each path is developed to allow the selection of the best candidate path. As multiple duplicate paths can be created using the branching planner, a strategy to remove these paths and greatly increase computation speed is discussed. Both simulation and hardware results are presented for validation.

When working with a complex AI or robotics system in a specific application, users often need to incorporate their special domain knowledge into the autonomous system. Such needs call for the ability to manage autonomy. However, managing autonomy can be a difficult task because the internal mechanisms and algorithms of the autonomous components may be beyond the users' understanding. We propose an approach where users manage autonomy indirectly by managing information provided to the intelligent system hierarchically at three different temporal scales: strategic, between-episodes, and within-episode. Information management tools at multiple temporal scales allow users to influence the autonomous behaviors of the system without the need for tedious direct/manual control. Information fed to the system can be in the forms of areas of focus, representations of task difficulty, and the amount of autonomy desired. We apply this approach to using an Unmanned Aerial Vehicle (UAV) to support Wilderness Search and Rescue (WiSAR). This dissertation presents autonomous algorithms/components and autonomy management tools/interfaces we designed at different temporal scales, and provides evidence that the approach improves the performance of the human-robot team and the experience of the human partner.

No description available.