An autonomous navigational system for a mobile robot was developed based on a Laser-Range-Finder-based path planning and navigational algorithms. The system was enhanced by incorporating collision avoidance algorithms using data from a sonar sensor array, and further improved by establishing two virtual regions in front of the robot for obstacle detection and avoidance. Several virtual detector bands with varying dimensions were also added to the sides of the robot to check for rotational clearance safety and to determine the direction of rotation. The autonomous navigational system was tested extensively under indoor environment. Test results showed that the system performed satisfactorily in navigating the mobile robot in three structured mazes under indoor conditions. / An artificial landmark localization algorithm was also developed to continuously record the positions of the robot whilst it was moving. The algorithm was tested on a grid layout of 6 m x 6 m. The performance of the artificial landmark localization technique was compared with odometric and inertial measurements obtained using a dead-reckoning method and a gyroscope-corrected dead-reckoning method. The artificial landmark localization method resulted in much smaller root mean square error (0.033 m) of position estimates compared to the other two methods (0.175 m and 0.135 m respectively).

Motion planning in the presence of obstacles deals with finding efficient paths from source to target which avoid hitting any of the obstacles. Applications include motion planning in robotics and designing efficient routing systems.Theoretical concepts from graph theory, topology, and computational geometry form the basis for some of the algorithms used in motion planning. The visibility graph is a standard model used in the study of motion planning.This thesis is a report of the research project undertaken to study visibility graphs and their applications to motion planning for certain geometric objects: polygons, line segments and points.The thesis consists of two parts.Theory: This part contains the details of topics from graph theory, topology, and computational geometry in motion planning. It also includes new algorithms which were developed as a part of this thesis.Implementation: This part describes a software system to implement the theory as an example of real applications. This software also includes many tools which help in studying visibility graphs. / Department of Computer Science

Pure reinforcement learning does not scale well to domains with many degrees of freedom and particularly to continuous domains. In this thesis, we introduce a hybrid method in which a symbolic planner constructs all approximate solution to a control problem.. Subsequently, a numerical optimisation algorithm is used to refine the qualitative plan into an operational policy. The method is demonstrated on the problem of learning a stable walking gait for a bipedal robot. The contributions of this thesis are as follows. Firstly, the thesis proposes a novel way to generate gait patterns by using a genetic algorithm to generate walking gaits for a humanoid robot using zero moment point as the stability criterion. This is validated on physical robot. Second, we propose an innovative generic learning method that utilises the trainer's domain knowledge about the task to accelerate learning and extend the capabilities of the learning algorithm. The proposed method, which takes advantage of domain knowledge and combines symbolic planning and learning to accelerate and reduce the search space of the learning problem, is tested on a bipedal humanoid robot learning to walk. Finally, it is shown that the extended capability of the learning algorithm handles high complexity learning tasks in the physical world with experimental verification on a physical robot.

No description available.

For mobile robots to autonomously accommodate dynamically changing navigation tasks in a goal-directed fashion, they must employ navigation plans. Any such plan must provide for the robot’s immediate and continuous need for guidance while remaining highly flexible in order to avoid costly computation each time the robot’s perception of the world changes. Due to the world’s uncertainties, creation and maintenance of navigation plans cannot involve arbitrarily complex processes, as the robot’s perception of the world will be in constant flux, requiring modifications to be made quickly if they are to be of any use. This work introduces Navigation Templates (or NaTs) which are building blocks for the construction and maintenance of rough navigation plans which capture the relationship that objects in the world have to the current navigation task. By encoding only the critical relationship between the objects in the world and the navigation task, a NaT-based navigation plan is highly flexible; allowing new constraints to be quickly incorporated into the plan and existing constraints to be updated or deleted from the plan. To satisfy the robot’s need for immediate local guidance, the NaTs forming the current navigation plan are passed to a transformation function. The transformation function analyzes the plan with respect to the robot’s current location to quickly determine (a few times a second) the locally preferred direction of travel. This dissertation presents NaTs and the transformation function as well as the needed support systems to demonstrate the usefulness of the technique for controlling the actions of a mobile robot operating in an uncertain world. ¹ This work was supported in part by a grant from the Jet Propulsion Laboratory under a contract from the National Aeronautics and Space Administration, and by a grant from the Naval Surface Weapons Center. / Ph. D.

The acquisition of capable robot manipulation skills is a critical prerequisite for the widespread deployment of robots in real-world environments, from household tasks to industrial applications. However, current robot manipulation systems remain limited in their ability to handle the diversity of objects, materials, and manipulation actions required in the real world. Data-driven methods have shown impressive results toward generalizing across a variety of problems, but existing approaches often require costly data collection using real robot platforms, hindering the scalability of skill acquisition. In this dissertation, we aim to push the limits of robotic manipulation task diversity by providing mechanisms to acquire new skills in a scalable manner. Achieving the "right" data with large quantity and high quality is of vital importance. We approach this problem by leveraging physics simulators. Different from commonly used rigid body simulators, we have customized simulators to support deformable objects with diverse materials and dynamics. Aerodynamics and fracture effects are also included to enable a wider range of manipulation actions such as blowing and cutting. With sophisticated system design, including proper representation selection and customized hardware design, the policies trained in simulation can be seamlessly applied to real robots. More specifically, this dissertation presents a series of works to address the challenges of diversity and scalability in robot manipulation skill acquisition. First, we introduce UMPNet, a universal policy network that can infer closed-loop action sequences for manipulating a wide range of articulated objects using only visual input. Second, we present DextAIRity, a system that leverages active airflow to enable safe and effective deformable object manipulation, expanding the repertoire of skills beyond traditional contact-based methods. Third, we describe RoboNinja, a cutting system for multi-material objects. With an interactive state estimator and an adaptive cutting policy, RoboNinja successfully removes the soft part of an object while preserving the rigid core.

by Law Kwok Ho Cedric. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2001. / Includes bibliographical references (leaves 106-112). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Motivation --- p.1 / Chapter 1.2 --- Related work --- p.4 / Chapter 1.3 --- Thesis overview --- p.5 / Chapter 2 --- Single wheel robot: Gyrover --- p.9 / Chapter 2.1 --- Background --- p.9 / Chapter 2.2 --- Robot concept --- p.11 / Chapter 2.3 --- System description --- p.14 / Chapter 2.4 --- Flywheel characteristics --- p.16 / Chapter 2.5 --- Control patterns --- p.20 / Chapter 3 --- Learning Control --- p.22 / Chapter 3.1 --- Motivation --- p.22 / Chapter 3.2 --- Cascade Neural Network with Kalman filtering --- p.24 / Chapter 3.3 --- Learning architecture --- p.27 / Chapter 3.4 --- Input space --- p.29 / Chapter 3.5 --- Model evaluation --- p.30 / Chapter 3.6 --- Training procedures --- p.35 / Chapter 4 --- Control Architecture --- p.38 / Chapter 4.1 --- Behavior-based approach --- p.38 / Chapter 4.1.1 --- Concept and applications --- p.39 / Chapter 4.1.2 --- Levels of competence --- p.44 / Chapter 4.2 --- Behavior-based control of Gyrover: architecture --- p.45 / Chapter 4.3 --- Behavior-based control of Gyrover: case studies --- p.50 / Chapter 4.3.1 --- Vertical balancing --- p.51 / Chapter 4.3.2 --- Tiltup motion --- p.52 / Chapter 4.4 --- Discussions --- p.53 / Chapter 5 --- Implement ation of Learning Control --- p.57 / Chapter 5.1 --- Validation --- p.57 / Chapter 5.1.1 --- Vertical balancing --- p.58 / Chapter 5.1.2 --- Tilt-up motion --- p.62 / Chapter 5.1.3 --- Discussions --- p.62 / Chapter 5.2 --- Implementation --- p.65 / Chapter 5.2.1 --- Vertical balanced motion --- p.65 / Chapter 5.2.2 --- Tilt-up motion --- p.68 / Chapter 5.3 --- Combined motion --- p.70 / Chapter 5.4 --- Discussions --- p.72 / Chapter 6 --- Shared Control --- p.74 / Chapter 6.1 --- Concept --- p.74 / Chapter 6.2 --- Schemes --- p.78 / Chapter 6.2.1 --- Switch mode --- p.79 / Chapter 6.2.2 --- Distributed mode --- p.79 / Chapter 6.2.3 --- Combined mode --- p.80 / Chapter 6.3 --- Shared control of Gyrover --- p.81 / Chapter 6.4 --- How to share --- p.83 / Chapter 6.5 --- Experimental study --- p.88 / Chapter 6.5.1 --- Heading control --- p.89 / Chapter 6.5.2 --- Straight path --- p.90 / Chapter 6.5.3 --- Circular path --- p.91 / Chapter 6.5.4 --- Point-to-point navigation --- p.94 / Chapter 6.6 --- Discussions --- p.95 / Chapter 7 --- Conclusion --- p.103 / Chapter 7.1 --- Contributions --- p.103 / Chapter 7.2 --- Future work --- p.104

Typescript (Photocopy) Includes paper co-authored by the author as attachment. Bibliography: leaves 133-140.

We study vertical jumping in a simple robot comprising an actuated mass spring arrangement. The actuator frequency and phase are systematically varied to find optimal performance. Optimal jumps occur above and below (but not at) the robot's resonant frequency f0. Two distinct jumping modes emerge: a simple jump which is optimal above f0 is achievable with a squat maneuver, and a peculiar stutter jump which is optimal below f0 is generated with a countermovement. A simple dynamical model reveals how optimal lift-off results from non-resonant transient dynamics.

This thesis presents a control system for a walking machine leg. The leg is representative of one of the six legs required for a proposed walking machine based on the geometry of the darkling beetle. Each of the three joints is controlled by a DC servo motor mounted to the base of the leg. The speed of the motors is controlled with pulse width modulation. Feedback of joint positions is accomplished with potentiometers mounted on the actual joints. A five-point path, forming a rectangle in the global coordinate system, is used as a skeleton of the path of movement. Desired times and accelerations from point to point are used to develop the path of movement, which smoothes corners and velocity transitions along the path. To create a model of the dynamics of each joint, a constant motor speed is output and the joint velocity and joint angle are recorded. From several trials at several different motor speeds, relationships between the joint velocity, joint angle, and motor speed can be found. This data is then least squares fit in two dimensions to give two second order functions. The first function uses the desired joint angle to calculate the variance from the mean joint velocity. This variance is then added to the desired joint velocity and is used in the second function to calculate the needed motor signal. Feedback control is accomplished using a PID control system. Because of the high level of noise in the feedback signal, a digital noise filter is used. Both moving average and linear regression techniques are examined. Performance of the system is measured by comparing the actual path in Cartesian coordinates to the desired path of movement. The RMS error is taken along the path, during the time frame of the ideal system. The maximum Cartesian error along the path is also used in evaluation. To determine suitable feedback gain combinations, several experiments are run and evaluated. Data is plotted and suitable values are chosen for the feedback gains based on their performance and sensitivity to change in performance. The performance of the leg is measured for a basic rectangular path, the basic path with a variation in step angle, and the basic path with a constant body velocity. / Graduation date: 1992