Alan G. Millard

Run-time detection of faults in autonomous mobile robots based on the comparison of simulated and real robot behaviour

This paper presents a novel approach to the run-time detection of faults in autonomous mobile robots, based on simulated predictions of real robot behaviour. We show that although simulation can be used to predict real robot behaviour, drift between simulation and reality occurs over time due to the reality gap. This necessitates periodic reinitialisation of the simulation to reduce false positives. Using a simple obstacle avoidance controller afflicted with partial motor failure, we show that selecting the length of this reinitialisation time period is non-trivial, and that there exists a trade-off between minimising drift and the ability to detect the presence of faults.

Towards Exogenous Fault Detection in Swarm Robotic Systems

It has long been assumed that swarm systems are robust, in the sense that the failure of individual robots will have little detrimental effect on a swarm’s overall collective behaviour. However, Bjerknes and Winfield [1] have recently shown that this is not always the case, particularly in the event of partial failures (such as motor failure). The reliability modelling in [1] shows that overall system reliability rapidly decreases with swarm size, therefore this is a problem that cannot simply be solved by adding more robots to the swarm. Instead, future large-scale swarm systems will need an active approach to dealing with failed individuals if they are to achieve a high level of fault tolerance.

Searching for pareto-optimal randomised algorithms

Randomised algorithms traditionally make stochastic decisions based on the result of sampling from a uniform probability distribution, such as the toss of a fair coin. In this paper, we relax this constraint, and investigate the potential benefits of allowing randomised algorithms to use non-uniform probability distributions. We show that the choice of probability distribution influences the non-functional properties of such algorithms, providing an avenue of optimisation to satisfy non-functional requirements. We use Multi-Objective Optimisation techniques in conjunction with Genetic Algorithms to investigate the possibility of trading-off non-functional properties, by searching the space of probability distributions. Using a randomised self-stabilising token circulation algorithm as a case study, we show that it is possible to find solutions that result in Pareto-optimal trade-offs between non-functional properties, such as self-stabilisation time, service time, and fairness.

An FPGA-based hardware-efficient fault-tolerant astrocyte-neuron network

The human brain is structured with the capacity to repair itself. This plasticity of the brain has motivated researchers to develop systems which have similar capabilities of fault tolerance and self-repair. Recent research findings have proven that interactions between astrocytes and neurons can actuate brain-like self-repair in a bidirectionally coupled astrocyte-neuron system. This paper presents a hardware realization of the bio-inspired self-repair architecture on an FPGA. We also introduce a reduced architecture for an FPGA-based hardware-efficient fault-tolerant system. This is based on the principle of retrograde signaling in an astrocyte-neuron network by simplifying the calcium dynamics within the astrocyte. The hardware optimized implementation shows more than a 90% decrease in hardware utilization and proves an efficient implementation for a large-scale astrocyte-neuron network. An Average spike rate of 0:027 spikes per clock cycle were observed for both the proposed models of astrocytes in the case of 100% partial fault.

The Psi Swarm: A Low-Cost Robotics Platform and Its Use in an Education Setting

The paper introduces the Psi Swarm robot, a platform developed to allow both affordable research in swarm robotics and versatility for teaching programming and robotics concepts. Motivated by the goals of reducing cost and construction complexity of existing swarm platforms, we have developed a trackable, sensor-rich and expandable platform which needs only a computer with internet browser to program. This paper outlines the design of the platform and the development of a tablet-computer based programming environment for the robot, intended to teach primary school aged children programming concepts.

Homeostatic Fault Tolerance in Spiking Neural Networks: A Dynamic Hardware Perspective

Fault tolerance is a remarkable feature of biological systems and their self-repair capability influence modern electronic systems. In this paper, we propose a novel plastic neural network model, which establishes homeostasis in a spiking neural network. Combined with this plasticity and the inspiration from inhibitory interneurons, we develop a fault-resilient robotic controller implemented on an FPGA establishing obstacle avoidance task. We demonstrate the proposed methodology on a spiking neural network implemented on Xilinx Artix-7 FPGA. The system is able to maintain stable firing (tolerance ±10%) with a loss of up to 75% of the original synaptic inputs to a neuron. Our repair mechanism has minimal hardware overhead with a tuning circuit (repair unit) which consumes only three slices/neuron for implementing a threshold voltage-based homeostatic fault-tolerant unit. The overall architecture has a minimal impact on power consumption and, therefore, supports scalable implementations. This paper opens a novel way of implementing the behavior of natural fault tolerant system in hardware establishing homeostatic self-repair behavior.

ARDebug: an augmented reality tool for analysing and debugging swarm robotic systems

Despite growing interest in collective robotics over the past few years, analysing and debugging the behaviour of swarm robotic systems remains a challenge due to the lack of appropriate tools. We present a solution to this problem—ARDebug: an open-source, cross-platform, and modular tool that allows the user to visualise the internal state of a robot swarm using graphical augmented reality techniques. In this paper we describe the key features of the software, the hardware required to support it, its implementation, and usage examples. ARDebug is specifically designed with adoption by other institutions in mind, and aims to provide an extensible tool that other researchers can easily integrate with their own experimental infrastructure.

The Need for Combining Implicit and Explicit Communication in Cooperative Robotic Systems

As the number of robots used in warehouses and manufacturing increases, so too does the need for robots to be able to manipulate objects, not only independently, but also in collaboration with humans and other robots. Our ability to effectively coordinate our actions with fellow humans encompasses several behaviours that are collectively referred to as joint action, and has inspired advances in human-robot interaction by leveraging our natural ability to interpret implicit cues. However, our capacity to efficiently coordinate on object manipulation tasks remains an advantageous process that is yet to be fully exploited in robotic applications. Humans achieve this form of coordination by combining implicit communication (where information is inferred) and explicit communication (direct communication through an established channel) in varying degrees according to the task at hand. Although these two forms of communication have previously been implemented in robotic systems, no system exists that integrates the two in a task-dependent adaptive manner. In this paper, we review existing work on joint action in human-robot interaction, and analyse the state-of-the-art in robot-robot interaction that could act as a foundation for future cooperative object manipulation approaches. We identify key mechanisms that must be developed in order for robots to collaborate more effectively, with other robots and humans, on object manipulation tasks in shared autonomy spaces.

Self-repairing Learning Rule for Spiking Astrocyte-Neuron Networks

In this paper a novel self-repairing learning rule is proposed which is a combination of the spike-timing-dependent plasticity (STDP) and Bienenstock, Cooper, and Munro (BCM) learning rules: in the derivation of this rule account is taken of the coupling of GABA interneurons to the tripartite synapse. The rule modulates the plasticity level by shifting the plasticity window, associated with STDP, up and down the vertical axis as a function of postsynaptic neural activity. Specifically when neurons are inactive, the window is shifted up the vertical axis (open) and as the postsynaptic neuron activity increases and, as learning progresses, the plasticity window moves down the vertical axis until learning ceases. Simulation results are presented which show that the proposed approach can still maintain the network performance even with a fault density approaching 80% and because the rule is implemented using a minimal computational overhead it has potential for large scale spiking neural networks in hardware.

Assessing Self-Repair on FPGAs with Biologically Realistic Astrocyte-Neuron Networks

This paper presents a hardware based implementation of a biologically-faithful astrocyte-based selfrepairing mechanism for Spiking Neural Networks. Spiking Astrocyte-neuron Networks (SANNs) are a new computing paradigm which capture the key mechanisms of how the human brain performs repairs. Using SANN in hardware affords the potential for realizing computing architecture that can self-repair. This paper demonstrates that Spiking Astrocyte Neural Network (SANN) in hardware have a resilience to significant levels of faults. The key novelty of the paper resides in implementing an SANN on FPGAs using fixed-point representation and demonstrating graceful performance degradation to different levels of injected faults via its self-repair capability. A fixed-point implementation of astrocyte, neurons and tripartite synapses are presented and compared against previous hardware floating-point and Matlab software implementations of SANN. All results are obtained from the SANN FPGA implementation and show how the reduced fixedpoint representation can maintain the biologically-realistic repair capability