Rapid fault-recovery strategies for resilient robot swarms

Lead Research Organisation: University of Southampton
Department Name: Sch of Electronics and Computer Sci

Abstract

Robots are increasingly becoming an important part of our day-to-day lives, automating tasks such as keeping our homes clean, and picking/packing our parcels at large warehouses. An aging population and the need to substitute human workers in dangerous and repetitive tasks have now resulted in new tasks on the horizon (e.g., in agriculture automation and environmental monitoring), requiring our robots to do more, to work in large-numbers as part of a swarm (a large team of robots), to coordinately sense and act over vast areas, and efficiently perform their mission. However, our robot swarms to date are unprepared for deployment; unable to deal with the inevitable damages and faults sustained during operation, they remain frail systems that cease functioning in difficult conditions. The goal of this project is to remedy this situation by developing algorithms for robot swarms to rapidly -- in no more than a few minutes -- recover from faults and damages sustained by robots of the swarm.

The existing fault-tolerant systems for robot swarms are limited. They are constrained to only diagnose faults anticipated a priori by the designer, which can hardly encompass all the possible scenarios a robot swarm may encounter while operating in complex environments for extended periods of time. The multitude of robots in a swarm and the large number of intricate ways they can interact with each other makes it difficult to predict potential faults and predefine corresponding recovery strategies; which may explain why none of the existing fault-detection and fault-diagnosis systems have been extended to provide fault-recovery mechanisms for robot swarms. Therefore, in order to design fault-tolerant algorithms for robot swarms, we need to move beyond the traditional approaches relying on fault-diagnosis information for fault recovery.

Fault recovery in a robot swarm may instead be formulated as an online behavior-adaptation process. With such an approach, the robots of the swarm adapt their behavior to sustained faults by learning via trial-and-error new compensatory behaviors that work despite the faults. However, the current approaches to learning new robot swarm behaviors are time-consuming, requiring several hours. Therefore, such approaches are inappropriate for behavior adaptation (learning new swarm behaviors) for rapid fault recovery.

Behavior adaptation for effective fault recovery requires the robot swarm to creatively and rapidly learn new compensatory swarm behaviors online, that work despite the sustained faults, effectively recovering the swarm from the faults. The proposal will address these requirements by investigating data-efficient machine learning techniques for rapid online behavior adaptation, guided by creatively and automatically generated intuitions -- evolved offline -- of working swarm behaviors. The resulting system would have a significant impact on long-term operations of robot swarms, and open up new and interesting applications for their deployment, such as the monitoring of large bodies of water for pollutants using a swarm of autonomous surface vehicles.

Planned Impact

The proposed project aims to develop algorithmic approaches for large-scale low-cost robot swarms to rapidly recover -- in no more than a few minutes -- from faults and damages sustained by individual robots of the collective. This will be achieved through the development of a novel combination of data-efficient machine-learning, with nature-inspired computing algorithms. Though close collaboration with project partners, the developed fault-recovery algorithms will be used to help improve the long-term autonomy of robot swarms, evaluated on a indoor swarm of mobile robots, and an outdoor swarm of aquatic surface drones. The project outcome is geared towards developing resilient robot swarms, working seamlessly around us, delivering the following economic and societal contributions.

Economic impact: A June 2014 report by the Intellectual Property Office indicates that the UK Government has identified Robotic and Autonomous Systems (RAS) as one of eight impactful technologies with the potential to propel the UK to future growth (goo.gl/YBSJ24). Developed RAS technologies are predicted to enable the conception of new products and services, a market of over £70 billion by 2025, disrupting existing markets as divergent as environmental monitoring, transport, manufacturing, and medicine (goo.gl/hkxDJY). Therefore, it is essential for the UK to be at the forefront of this research area, to capture value as the markets disrupt and change.

Autonomous operation will be an essential feature of the next generation of robot swarms, allowing the robots to continue functioning despite faults resulting from common wear and tear of their functional parts, and unexpected changes in their operational environments. This ability increases the usefulness of the robots, and extends the amount of time they can continue operating without human intervention, thus providing a substantial economic advantage. For instance, industries deploying robot swarms do not expect their system to grind to a halt, every time the robots encounter an unanticipated situation, and individual robots suffer damages. They would preferably avoid disruptions in their regular operations, and the hefty start-up costs required to provide existing robot swarms with carefully controlled working environments, if that were even possible (e.g., robot swarms operating outdoors). The proposed research will contribute towards the development of long-term autonomous robot swarms, ideal for such scenarios.

One of the potentially impactful real-world application scenario for robot swarms is that of autonomous surface vehicles (ASVs) in rivers, lakes and marine environments. In such environments, robots are required to operate over vast areas, with minimal human intervention. ASV robots have been commercialized, and widely used in commerce, industry and military applications; such an established and active market for the platform now opens up impactful avenues for applications with ASV swarms. Finally, a number of industry players in the UK have keen interests in this area. Engagement with these players will be fostered by industrially relevant (out of the lab) robot swarm case-studies, developed in collaboration with the Southampton Marine and Maritime Institute, MBDA-UK Ltd. and ASV Global.

Societal impact: The robot swarms envisioned in this project, capable of extended autonomy, also have the potential to directly impact important scientific endeavors, beyond the robotics community. In these endeavors, robots will be employed as a fundamental data-gathering tool, allowing a swarm of robots to provide us with a rich and continual stream of high-resolution data on observed environmental phenomena. For instance, a large swarm of aquatic surface drones fitted with Passive Acoustic Monitoring sensors, may intelligently monitor a water-body for the presence of marine mammals, providing valuable data for the regulation of shipping lanes.
 
Description The existing fault-tolerant systems for robot swarms are limited. They are constrained to only diagnose faults anticipated a priori by the designer, which can hardly encompass all the possible scenarios a robot swarm may encounter while operating in complex environments for extended periods. The large multitude of robots in a typical swarm and the large number of intricate ways they can interact with each other makes it difficult to predict potential faults and predefine corresponding recovery strategies; which may explain why none of the existing fault-detection and fault-diagnosis systems have been extended to provide fault-recovery mechanisms for robot swarms.

In this project, my team and I developed novel fault-tolerant algorithms for robot swarms, moving well beyond the traditional approaches that rely on fault-diagnosis information for fault recovery. Fault recovery in a robot swarm is formulated as an online behaviour-adaptation process. Robots of our swarm adapt their behaviour to sustained faults by learning via trial-and-error new compensatory behaviors that work despite the faults. To achieve this goal, we first developed large repertoires of compensatory behaviours for the swarm. For our experiments, over 250 unique faults were injected into the robots of the swarm, in the sensors, actuators and disruptions to to the environment. Interestingly, with our proposed Quality-Environment-Diversity algorithm, the swarm sustained a median 2- to 3-fold reduction in the impact of injected faults. In a further study, to improve the recovery time of the swarm, we proposed the Swarm Map-based Optimisation algorithm. The developed algorithm was successful in fault recovery; improving performance up to 80% compared to the performance at the time of fault injection within at most 30 trials.

In conclusion, the grant has allowed me to successfully lay the algorithmic foundations for rapid fault-recovery in robot swarms. The applications of this research are now being pursued in environmental monitoring applications, such as precision forestry.
Exploitation Route Robot swarms are groups of robots that each act autonomously based on only local perception and coordination with neighbouring robots. While current swarm implementations can be large in size (e.g., 1,000 robots), they are typically constrained to working in highly controlled indoor environments. Moreover, a common property of swarms is the underlying assumption that the robots act in close proximity of each other (e.g., 10 robot body lengths apart), and typically employ uninterrupted, situated, close-range communication for coordination. Many real-world applications, including environmental monitoring and precision forestry, however, require scalable groups of robots to act jointly over large distances (e.g., 1,000 robot body lengths apart), rendering the use of dense swarms impractical. Using a dense swarm for such applications would be invasive to the environment and unrealistic in terms of mission deployment, maintenance and post-mission recovery.

To address this problem, I propose the sparse swarm concept where group of robots interact while (i) not being in close proximity to each other, and/or (ii) it is not possible for information to rapidly propagate within the group. Sparse swarms are particularly relevant in application scenarios where the robots are operating 100s meters apart under sporadic low-bandwidth communication constraints. Due to the large distances between robots of the swarm, they have to exhibit a high-degree of autonomy, and capability to quickly adapt to changing environments. Therefore, this builds on my earlier work done in my New Investigator Award to develop rapid behaviour-adaptation algorithms for resilient robots. I am now investigating the applications of sparse swarms in terrestrial, marine and aerial real-world environments. For example, in the terrestrial domain I am investigating the high-impact scenario of forest monitoring. In this scenario, a group of rovers is required to traverse, and monitor, large forest environments. The research challenges addressed here comprise the rover hardware development, low-cost navigation strategies for small-sized rovers and communication-constrained coordination algorithms for the swarm.
Sectors Aerospace, Defence and Marine,Agriculture, Food and Drink,Digital/Communication/Information Technologies (including Software),Electronics,Energy,Environment
 
Description The project was a success, my team and I disseminated our research findings through nine peer-reviewed publications in leading international conferences and journals. The foundational nature of the research in this project has limited the non-academic impact. Importantly, in the course of the project, my team and I have designed and developed the following two innovative and creative open-source tools to assist with research on robot swarms in extreme environments: i. Developed, to the best of my knowledge, the first high-fidelity open-source simulator of autonomous surface vehicles in marine environments in sea states ranging from calm to stormy. Innovatively used the small size of deployed ASV assets to improve the performance of the simulator so that it can run on low-cost embedded platforms. The GitHub repository of the simulator has so far received over 10 stars. ii. Developed an extensive open-source dataset of over 100K forest images captured under different weather conditions and times of day. A mobile platform was designed to easily capture colour and depth images from the required low-viewpoint. To the best of my knowledge, this is the only such dataset available for forest environments. It has garnered interest from roboticist and ecologist alike, and has so far received over 540 views and 175 downloads on Zenodo.
First Year Of Impact 2023
Sector Education,Environment
 
Title ASVLite: a high-performance simulator for autonomous surface vehicles 
Description The energy of ocean waves is the key distinguishing factor of marine environments compared to other aquatic environments such as lakes and rivers. Waves significantly affect the dynamics of marine vehicles; hence it is imperative to consider the dynamics of vehicles in waves when developing efficient control strategies for autonomous surface vehicles (ASVs). However, most marine simulators available open-source either exclude dynamics of vehicles in waves or use methods with high computational overhead. This paper presents ASVLite, a computationally efficient ASV simulator that uses frequency domain analysis for wave force computation. ASVLite is suitable for applications requiring low computational overhead and high run-time performance. Our tests on a Raspberry Pi 2 and a mid-range desktop computer show that the simulator has a high run-time performance to efficiently simulate irregular waves with a component wave count of up to 260 and large-scale swarms of up to 500 ASVs. 
Type Of Material Computer model/algorithm 
Year Produced 2021 
Provided To Others? Yes  
Impact As the simulation software has been recently released, no notable impact as yet is there to report. 
URL https://github.com/resilient-swarms/ASVLite
 
Title Low-viewpoint forest depth dataset for sparse rover swarms 
Description Rapid progress in embedded computing hardware increasingly enables on-board image processing on small robots. This development opens the path to replacing costly sensors with sophisticated computer vision techniques. A case in point is the prediction of scene depth information from a monocular camera for autonomous navigation. Motivated by the aim to develop a robot swarm suitable for sensing, monitoring, and search applications in forests, we have collected a set of RGB images and corresponding depth maps. Over 100000 RGB/depth image pairs were recorded with a custom rig from the perspective of a small ground rover moving through a forest. Taken under different weather and lighting conditions, the images include scenes with grass, bushes, standing and fallen trees, tree branches, leaves, and dirt. In addition GPS, IMU, and wheel encoder data were recorded. From the calibrated, synchronized, aligned and timestamped frames about 9700 image-depth map pairs were selected for sharpness and variety. We provide this dataset to the community to fill a need identified in our own research and hope it will accelerate progress in robots navigating the challenging forest environment. This paper describes our custom hardware and methodology to collect the data, subsequent processing and quality of the data, and how to access it. 
Type Of Material Database/Collection of data 
Year Produced 2020 
Provided To Others? Yes  
Impact As the simulation software has been recently released, no notable impact as yet is there to report. 
URL https://zenodo.org/record/3945526#.Yiimj3rP02w