Browsing by Subject "Autonomous navigation"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Robot Motion Planning for Tracking and Capturing Adversarial, Cooperative and Independent Targets(2015-09) Karnad, NikhilThe last ten years have seen robots become smaller, lighter and more maneuverable than ever before, paving the way for their use in the service sector, and even in our homes. A key aspect that elevates a network of robots over a bed of sensors is the ability of individual units to autonomously navigate, forming a sensor-actuator network. These added degrees of freedom demand that we revisit existing motion planning and navigation algorithms and extend them to be able to better contribute to the services they are designed to provide. In deploying autonomous robots away from laboratory-controlled settings and in to real-world environments, care must be taken because they share their space with other autonomous agents. For instance, a typical home robot has to account for people as well as their pets. Imparting a fair degree of autonomy to robot navigation tasks thus requires reasoning about the mobility of other entities in the path-planning process. This is not a trivial task because such agents can not often be easily characterized or quantified, e.g. human navigation depends on many factors including the perception of hazards, experience from long-term memory, even whim. In this thesis, we take a deep look at the role of the information available to path-planning algorithms in the context of the mobile target-tracking problem. We start with the problem formulation (Chapter 2), in which robots need to keep track of a single mobile target for as long a duration of time as possible. In the game-theoretic sense, the information available to the target has a different characteristic than the information available to the robots. For example, target behavior varies from adversarial ones in defense applications to cooperative ones in service applications. In the chapters that follow, we present the main contributions of this thesis with this role of information during the path-planning process being the common thread. Traditional surveillance applications assume the worst-case – the target is an adversary actively trying to escape from the robots. We model this strategic interaction with the theory of pursuit-evasion games, wherein a robot pursues a target, which in turn tries to evade it. We study a mathematical variant of the Lion and Man game in the presence of an obstacle, and show that the final outcome of whether the lion can track down the man depends, in closed form, on the initial conditions (Chapter 3). We derive optimal player strategies that work regardless of what the other player does, thus providing the worst-case guarantees that such applications demand. At the opposite end of the spectrum, there exist service applications where the target’s trajectory is known to the robots before they need to move. Our motivating example is mobile telepresence, or virtual presence – an emerging technology that enables people to participate in environments they are not physically present in. Specifically, we present a navigation algorithm for a first-of-its-kind person-following robot with telepresence and monitoring applications (Chapter 4). We require that in the presence of another person, the robot should ensure a safe distance by following that person’s path without losing track. We present a heuristic planning approach that accounts for both non-holonomic constraints and trajectory smoothness. Further, a control-theoretic implementation is provided. Targets that navigate autonomously usually do so with their own intentions. While it is uncertain how they navigate, their behavior may neither be adversarial, nor completely known to the robots in advance. We present a novel data-driven probabilistic mobility model that can be used by path planners to reason about the uncertainties in decisions made by an individual who is navigating in an indoor environment (Chapter 6). We show that it is possible to preserve long temporal histories over an abstracted representation of the environment, which helps predict future mobility of the target better than previous approaches. We present a multi-robot planning algorithm that builds off of this model. Although our algorithm is designed for long-term planning and offline solution, we are able to execute the robot paths in real-time, and demonstrate extended utility with simulations. We discuss the architecture of a complete system implementation for the telepresence application, including both hardware design and software development (Chapter 8). With an increasing aging population in the U.S., it is our belief that such a system would become relevant in the near future, e.g., to assisted living facilities for purposes of healthcare monitoring.Item Six Degree Of Freedom Navigation Using X-Ray Pulsar Signals(2019-08) Runnels, JoelNavigation in deep space, far away from Earth, is an ongoing challenge and research topic. While spacecraft near Earth have a number of readily available methods for navigation (including GPS and radio ranging), far away from Earth it is more challenging for spacecraft to determine their position. In the absence of external reference objects that can be used to estimate position (for instance nearby planetary objects), the current state-of-the-art for navigation in space relies on NASA's Deep Space Network to provide Earth-based position measurements of the spacecraft. This means of navigation suffers from limitations, including limited availability, high cost, and decreased accuracy far from Earth. Consequently, alternative means of navigation are of interest. X-ray navigation, or XNAV is a proposed means by which spacecraft can navigate using signals generated by astrophysical signal sources. In particular, x-ray pulsars have been proposed as a naturally occurring signal source which could be used to generate a position, navigation and timing (PNT) solution in space. The basic concept in XNAV is that a spacecraft can compute a PNT solution based on time of arrival (TOA) measurement of signals from x-ray pulsars. Some x-ray pulsars, in particular millisecond pulsars, have extremely precise timing characteristics, with timing stability comparable to modern atomic clocks. If the TOA of signals from several millisecond pulsars could be measured, these TOAs could be used to compute a PNT solution for the spacecraft. The basic concept of XNAV is somewhat analogous to GPS, in that the position of the user is determined by measuring multiple signal TOAs generated by sources with precisely known timing characteristics. While this technique has been proposed numerous times in literature, there are still several implementation challenges which must be overcome in order for XNAV to become a viable navigation technology. In this dissertation, we address some of the major challenges associated with implementation of XNAV. The first challenge addressed in this dissertation is the development of a method of determining the signal time of arrival based on measurements of x-ray photon arrival times. This challenge is at the heart of any XNAV implementation, because in order to use pulsar signals as PNT signals, the time-difference of arrival of the signal must be measured. The estimation of time-difference of arrival from pulsar signals is complicated by the fact that pulsar signals are incredibly weak, resulting in a signal-to-noise ratio near zero. In this dissertation, we develop a recursive algorithm which estimates the time-difference of arrival of a pulsar signal which is based upon adaptive filtering techniques. The second challenge addressed in this dissertation is the problem of data association. Photons measured by an x-ray detector in space have no way of knowing with certainty the origin of the photons. The presence of the uniform x-ray background results in background photons diluting an already extremely weak signal. If the detector's attitude is known, then the attitude may be used to determine which photons are likely to have originated from a signal source of interest. However, the reliance upon attitude to correctly associate the photons with the correct signal source causes the position and attitude estimates to be coupled. In this dissertation, we present an algorithm which addresses this coupling of the attitude and PNT solutions for the XNAV problem. A joint six degree-of-freedom position and attitude estimator is developed based on the joint probabilistic data association filter. We further demonstrate the effects of attitude uncertainty on the accuracy of the PNT solution using Monte Carlo simulations.