Jonathan M. Aitken

Control theory for principled heap sizing

We propose a new, principled approach to adaptive heap sizing based on control theory. We review current state-of-the-art heap sizing mechanisms, as deployed in Jikes RVM and HotSpot. We then formulate heap sizing as a control problem, apply and tune a standard controller algorithm, and evaluate its performance on a set of well-known benchmarks. We find our controller adapts the heap size more responsively than existing mechanisms. This responsiveness allows tighter virtual machine memory footprints while preserving target application throughput, which is ideal for both embedded and utility computing domains. In short, we argue that formal, systematic approaches to memory management should be replacing ad-hoc heuristics as the discipline matures. Control-theoretic heap sizing is one such systematic approach.

Adaptation of System Configuration under the Robot Operating System

Abstract This paper lays down the foundations of developing a reconfigurable control system within the Robot Operating System (ROS) for autonomous robots. The essential components of robots are programmed under a ROS system. A formal model is defined as a tripartite graph to represent the robots functional architecture. ROS systems are then generalised to component libraries for any ROS architecture and an abstract model as a “system graph” is introduced. Orthogonality of a library and a system graph is defined and redundancy levels of robot components are studied for maintaining full functionality of the robot by automated reconfiguration in face of hardware malfunction. This allows AI planning tools, such as Planning Domain Definition Language (PDDL), to compute permissible reconfigurations. We present an example of a pair of robotic arms which requires reconfiguration of the underlying control system in order to retain the capability to carry out a task.

A model based design framework for safety verification of a semi-autonomous inspection drone

In this paper, we present a model based design approach to the development of a semi-autonomous control system for an inspection drone. The system is tasked with maintaining a set distance from the target being inspected and a constant relative pose, allowing the operator to manoeuvre the drone around the target with ease. It is essential that the robustness of the autonomous behaviour be thoroughly verified prior to actual implementation, as this will involve the flight of a large multi-rotor drone in close proximity to a solid structure. By utilising the Robotic Operating System to communicate between the autonomous controller and the drone, the same Simulink model can be used for numerical coverage testing, high fidelity simulation, offboard execution and final executable deployment.

A case for dynamic risk assessment in NEC systems of systems

The level of complexity in Systems of Systems is increasing as more complex functionality emerges from the interaction of individual components. As networks become more complex it becomes more difficult for an individual to identify potential safety risks. We know, from previous accidents, that poor understanding of networks can be dangerous. In this paper, we demonstrate the potential value of incorporating a process to identify risks in a deployed network, focusing on factors concerned with the interaction of this process with a user, and the potential for new hazards.

ROBO-GUIDE: Towards Safe, Reliable, Trustworthy, and Natural Behaviours in Robotic Assistants

In this paper we describe a novel scenario, whereby an assistive robot is required to use a lift, and results from a preliminary investigation into floor determination using readily-available information. The aim being to create an assistive robot that can naturally integrate into existing infrastructure.

Observer/Kalman Filter Identification With Wavelets

Understanding the dynamic characteristics of the target system is a fundamentally important step in designing reliable closed loop control systems. One method for identifying linear models from the target uses the observer/Kalman filter identification/eigensystem realization algorithm (OKID/ERA) combination. This focuses on time domain representations of plant input and output data. The wavelet transform is capable of providing an efficient mixed time-frequency domain representation of time domain data. We have investigated if this would give any benefit for OKID/ERA. We show how to apply the wavelet transform within the OKID creating a new wavelet-OKID/ERA technique and then compare its performance, using common discrete wavelet families, against a standard procedure. Results indicate that the potential attractiveness of the wavelet approach do not translate into a practical reality.

Deep Parameter Optimisation for Face Detection Using the Viola-Jones Algorithm in OpenCV

OpenCV is a commonly used computer vision library containing a wide variety of algorithms for the AI community. This paper uses deep parameter optimisation to investigate improvements to face detection using the Viola-Jones algorithm in OpenCV, allowing a trade-off between execution time and classification accuracy. Our results show that execution time can be decreased by 48 % if a 1.80 % classification inaccuracy is permitted (compared to 1.04 % classification inaccuracy of the original, unmodified algorithm). Further execution time savings are possible depending on the degree of inaccuracy deemed acceptable by the user.

Assessing Graphical Robot Aids for Interactive Co-working

The shift towards more collaborative working between humans and robots increases the need for improved interfaces. Alongside robust measures to ensure safety and task performance, humans need to gain the confidence in robot co-operators to enable true collaboration. This research investigates how graphical signage can support human–robot co-working, with the intention of increased productivity. Participants are required to co-work with a KUKA iiwa lightweight manipulator on a manufacturing task. The three conditions in the experiment differ in the signage presented to the participants—signage relevant to the task, irrelevant to the task, or no signage. A change between three conditions is expected in anxiety and negative attitudes towards robots; error rate; response time; and participants’ complacency, suggested by facial expressions. In addition to understanding how graphical languages can support human–robot co-working, this study provides a basis for further collaborative research to explore human–robot co-working in more detail.

Floor determination in the operation of a lift by a mobile guide robot

Robotic assistants operating in multi-floor buildings are required to use lifts to transition between floors. To reduce the need for environments to be tailored to suit robots, and to make robot assistants more applicable, it is desirable that they should make use of existing navigational cues and interfaces designed for human users. In this paper, we examine the scenario whereby a guide robot uses a lift to transition between floors in a building. We describe an experiment into combining multiple data sources, available to a typical robot with simple sensors, to determine which floor of the building it is on. We show the robustness of this approach to realistic scenarios in a busy working environment.

A risk modelling approach for a Communicating System of Systems

Communicating Systems of Systems provide new methods for developing flexibility and functionality. However, these systems contain added layers of complexity due to the emergent behaviours present in the network. In order to understand the safety implications of operating these SoS this paper develops a new risk modelling technique. Building on information contained within the MODAF model of the SoS this technique provides a structure to enable the analysis of risk.