Impact case study database

It is estimated that by 2022, the number of connected electronics worldwide will hit 29 billion, of which around 18 billion will occupy the burgeoning internet of things (IoT) subcategory [S1].

Surrey’s research and development of novel methods for supporting transactions between and collaborations amongst agents in a distributed ecosystem of services underpinned the decision by Dr Alexandros Marinos – a former Surrey PhD student and PDRA – to found the start-up company Balena (formerly RuleMotion) in 2011 with the aim of commercialising research outputs from Surrey’s Digital Ecosystem programme. Marinos was supported in this venture by Krause and Moschoyiannis, who together aimed to use Surrey’s expertise to make managing IoT devices as straightforward as configuring PC peripherals. Both Krause and Moschoyiannis continue as advisors to this company, with Krause also being a shareholder.

Impact on Balena’s product offering

The conceptual basis, and many of the technical outputs from Surrey’s Digital Ecosystems research group, are deeply embedded in the products and ethos of Balena. Balena’s API platform, open-sourced as “S3” [S2, S3] in its entirety, is a direct descendant of the work reported in [R2, R3]. SBVR is still used in production, and the system still compiles to SQL as described in these outputs [S4]. Founder and CEO Marinos states, “<...> it remains central to our core business and has been a key facilitator of our agile way of working.” [S2]

Balena’s first product, launched in 2012, was a syntax-directed editor for SBVR that could compile structured natural language models to SQL [R2]; this remains at the core of the Balena API. The rebranding of the original company to “Balena.io” in late 2013 coincided with the full-scale launching of their support for service management in fleets of IoT devices, which exploited the service composition and transactional research of [R1, R5, R6].

Balena as a catalyst for innovation

While the Balena product stack has been expanded significantly during the company's life, the Balena API remains at the heart of their IoT Infrastructure. Our research work [R2, R3, R4] was amongst the first to use “RESTful” ways of working with connected resources, and this remains at the heart of the Balena way of working. The use of a restricted set of verbs to interact with resources (including devices, applications, users, and more) provides a unified interface with which to interact with this diversity of resources. This way of working has almost universally replaced the baroque service-oriented architectures prevalent at the start of the new millennium. However, what made, and still makes, the API approach scalable and easy to use was the embedding of REST into a structured natural language interface using SBVR and the automated mapping into an SQL persistence layer [R2, R3]. This means that users of Balena do not need specialised knowledge of the service-oriented architecture and database technology in order to set up and manage a fleet of interacting resources [S4].

This ease of setting up the management of a fleet of resources is an aspect of Balena that is repeatedly mentioned as a primary reason for moving to Balena by innovative IoT businesses [S5, S6, S7]. The robust transaction model [R4, R5, R6] is less visible to the SME end-users but is recognised in terms of the quality of service they can provide [S5]. The concept of forward recovery formalised in [R5] is particularly important as it enables dynamic updating of micro-services to be facilitated without loss of service. This is particularly important for safety-critical examples, such as when Balena is used to manage fleets of drones [S5], with even the updating of a complete software stack having been demonstrated mid-flight. [S5] also reinforces how Balena enabled Skycatch to rapidly respond to a first-time request for a fleet of several hundred of their surveillance drones, scaling from the operation of small numbers to fleets of potentially thousands of drones simply through integrating Balena into their product stack.

Ease of use is especially illustrated in [S6], where the use of Balena was used by Figure Devices to take a prototype idea developed in a maker space into production support of hundreds of instances.

The importance of being able to support live updates of software [R4, R6] without “bricking” devices (making them unusable for a period) was a key driver behind OpenROV’s migration to Balena [S7].

However, Balena is not just a key enabler for innovation-driven SMEs. The use of Balena to support one of Europe’s biggest IoT hackathons and subsequent use for Bosch’s own products is described in [S8]. Again, the Balena dashboard features here; a core part of Balena that originated in Surrey’s research programme [R2, R3].

The appeal to such a broad customer base has led to an acceleration of demand for Balena; across all their customers, the number of devices supported by Balena’s infrastructure grew from c15,000 in January 2018 to over 70,000 in May 2019.

Growth of Balena

Three years after moving the company’s base to Washington, Seattle in 2013, it managed to attract $9 million in venture capital from Threshold Ventures, GE Ventures, Ericsson, and Aspect Ventures. With a subsequent $5 million in funding, Balena geared up for expansion, flush with $14.4 million raised in a series B round led by OpenView and existing investors. The infusion has fuelled its workforce's growth, distributed across its Seattle headquarters, satellite offices in London & Athens, and globally. The number of devices on Balena’s platform is growing at a rate six times year-over-year, and that customer acquisition is on the upswing. Balena now counts Bosch, Booster, Allscripts, Tapp, Zume Pizza, OpenDoor, Sonder, smart home systems company Dwelo, smart roof start-up Sense, HVAC monitoring and controls provider Enerbrain, and smart building controls developer Mount Kelvin among its customer base [S1].

Balena has now raised a total of $31.4 million in funding since April 2015 [S10]. The last two years, especially, have seen accelerating growth in Annual Recurring Revenue, from just under $500,000 in January 2018 to $1.9 million in May 2019. The company has grown from 5 employees in 2013 to over 100 employees, distributed globally [S9, S10].

Modern engineering design problems need to meet multiple, often conflicting, objectives. Research work undertaken at Surrey has a wide range of applicability but has demonstrated impact in three specific areas:

1. Optimisation of hybrid vehicle energy management controllers;

2. Optimisation of vehicle dynamics in various driving scenarios;

3. Making available a comprehensive, general-purpose platform for evolutionary multi-objective optimisation under an open-source licence.

Hybrid vehicles, with both electric and internal combustion power sources are gaining popularity as a means of reducing greenhouse gas emissions. Approximately 240,000 of these vehicles have been registered in the UK alone since 2012.

Hybrid vehicle energy management has, in the past, been seen as simply a matter of minimising fuel consumption. However, more objectives need to be taken into account if we are to optimise the overall reduction in environmental impact and take into account the overall driving experience. Whilst CO₂ emissions are closely correlated with fuel consumption, the emission of other greenhouse gases is associated with other phases of the internal combustion engine running cycle; emissions of nitrous oxides, for example, mostly occur when the engine is on, and the catalytic converter has not yet reached its operating temperature. In addition, the control strategy needs to minimise the usage of the engine within urban environments. However, recent studies have shown that hybrid vehicles can exhibit higher pollution levels in use than those claimed by the manufacturers [S1]. As a result, the overall assessment of the environmental impact of hybrid vehicles is now recognised as vitally important to avoid a backlash against their usage [S2]. For example, drivers may alter their driving practices to optimise their own preferences for noise reduction and charging frequency and this, in turn, can negatively impact vehicle emissions [S2]. Thus, it is important for the controller to optimise against a broader range of objectives to ensure a satisfactory driving experience is obtained whilst still meeting the pollution reduction objectives.

We therefore worked with Honda Research and Development to demonstrate that our research work can optimise against seven objectives: fuel consumption; battery charge/discharge cycling; engine operation changes; engine emissions; perceived engine noise; engine usage in urban operation; average battery state of charge level. This work showed that our “RVEA” algorithm [R3] provided an effective balance between convergence and diversity of solutions, supporting an approach that achieved the objectives preferred by the decision-maker [R4]. This positive experience has motivated the continued usage of MOEAs at Honda Research & Development Europe.

{Text removed for publication}

Vehicle stability at higher speeds is another important and challenging factor that must be addressed in vehicle design. For both safety and comfort, the vehicle must remain stable, predictable and controllable under a wide variety of road surface conditions and maintenance states. The vehicle behaviour is mainly influenced by the interaction of suspension components, the steering subsystem and tyre characteristics. This leads to a high number of adjustment factors and results in many possible variations depending on the vehicle targets. The trade-off between vehicle dynamics, ride comfort and road holding capability makes the definition of the perfect chassis setting challenging. This is normally tested using pre-production prototypes but testing and corrections to the design at this phase are both very expensive, usually requiring a number of redesign-test cycles. Honda R&D Germany has developed a simulation-based optimisation and evaluation approach that both reduces the cost and enhances the effectiveness of this design phase [S4]. Our algorithms are providing an important contribution to this technology. Honda R&D has been using our optimisation platform PlatEMO [R5] since 2019. Although PlatEMO contains a broad range of candidate algorithms, in an independent evaluation our “KnEA” algorithm was shown to be amongst the first class of performers on benchmark 7-objective problems [S5]. {Text removed for publication}

Our generation of impact is not just focused on Honda. All the algorithms, together with reports on their evaluation against real-world problems, are in the public domain. In addition, the evolutionary multi-objective optimisation platform PlatEMO [R4] (co-developed with the Institute of Bioinspired Intelligence and Mining Knowledge of Anhui University, China) is available free of charge to academia and industry [S6]. The importance of PlatEMO in building external confidence in the claims of the research community has been recognised by over 1,200 scientists from 40 countries who have (co-)authored papers citing it. In addition, PlatEMO is now (December 2020) in use at 22 industrial and public sector organisations, including Honda, Jaguar, Huawei, Ericsson, Diehl Aviation (Germany), AIST (Japan), Caterpillar Propulsion AB (Sweden) [S7].

The impact is concerned with the introduction of verifiability into electronic voting: its use in real elections in Australia and the UK; and influence on policy in the UK.

Australia: For the Victorian State Parliamentary Election in November 2014, the Victorian Electoral Commission (VEC) desired to ensure accessibility to a secret ballot for voters who might otherwise be disenfranchised, but security was a key consideration. VEC chose the University of Surrey’s verifiable vVote system [R3, R4, R5] as they felt it offered the most trustworthy solution [S4]. Voters in the State of Victoria who were visually impaired, mobility impaired, or had a language other than English, were able to cast binding ballots electronically using the Surrey vVote system across the State of Victoria. The system was also made available to all voters resident in the UK who voted at Victoria House, London. The team from Surrey led by Schneider worked with VEC to adapt the vVote system to their specific requirements, providing particular expertise on the back end which managed all the cryptography and the capture and handling of the votes. A feedback survey established that voter satisfaction was high.

The VEC reported “This was comparable to, or better than the findings in 2010, and pleasing considering the new steps introduced with the verification processes in 2014. Of the open-ended questions asked of the London electors, almost all answers were positive about the system. Overall, this is an important finding that shows independent verification is not an impost on electors when voting electronically ” [S2].

The then Director of Electronic Voting at VEC, Craig Burton, reported that “the success of the deployment swung the discussion [see S3] and approach to e-voting across the various Australian election commissions, by demonstrating that verifiable voting was technically feasible ” and that “the system has embedded the notion of verifiability […] it is now understood across the industry as a key security feature.” Indeed, many current competitor systems now claim to incorporate forms of verifiability. With respect to improving inclusion of potentially marginalised groups, Burton reported that he “trialled the system with profoundly physically disabled people”, and that “access to this system for them, were it legislated, would represent the first time in many disabled people’s lives where they could participate privately and independently in an election. For the profoundly physically disabled this may be one of the very few acts they can perform on their own at all. ” [S4].

UK: In the UK statutory elections are paper-based. However, organisations (e.g., trades unions, building societies, professional organisations, political parties, student unions) also run ballots. Civica Election Services (CES; formerly Electoral Reform Services Ltd) are the UK market leader in this sector with over 90% market share. In recent years, remote electronic voting has become popular to provide convenience for those entitled to vote, as well as allowing for cost saving and rapid elections. In the period October 2018-September 2019, we developed the approach [R6] to introduce verifiability into the CES existing internet voting system. This was used on binding ballots with clients, including an election for the Royal College of Nursing in July 2019 with an electorate of 40,000. Approximately 40% of votes cast used the verification system to check their vote. Another example is the ballot with the College of Podiatry in October 2019. Philip Wright, CES Technical Director, reported “the experience of having run verifiable ballots has demonstrated to the business that this can be done in a way that is commercially viable and attractive to customers […] it is clear that verifiability plays a key role.” He further comments that it has “enhanced our offering and market position without negatively impacting on the use of our core product.” [S5]

UK Policy – Industrial Actions Balloting: Schneider co-authored the technical annex for Sir Ken Knight’s Independent Review on Electronic Balloting for Industrial Action [S6] commissioned by the Department for Business, Energy and Industrial Strategy (BEIS), and made a direct impact on its policy recommendations [S7]. The annex together with Schneider’s involvement in the round table discussions and a fact-finding trip with Sir Ken to Cybernetica in Estonia “led directly to one of the key recommendations in [Sir Ken’s] report that verifiability should be included in electronic balloting systems when used for industrial action ballots” [S7] for consideration by UK Government developing policy on remote, electronic voting. Schneider also chairs the Institution of Engineering and Technology (IET) Working Group on Electronic Voting which includes members from industry, academia, government from UK and internationally, whose report ‘Internet Voting for the UK’ (October 2020) [S8] includes verifiability as a key recommendation.