Impact case study database

Google’s Director of Engineering for Android Gaming and Graphics stated: “The cumulative effect of this impact on the Android and Chrome platforms is an improvement in graphics driver reliability and security for billions of users around the world. [...] Having something like GraphicsFuzz to aid in the discovery of vulnerabilities and issues can save companies like Google billions of dollars.” Savage also calls out the relationship between this impact and the underpinning research: “The impact is a direct result of the first-class research undertaken at Imperial College London that lies behind the GraphicsFuzz approach.” [I1]. It is estimated that there are more than 2.5 billion active Android devices worldwide, and that 66% of all internet users use the Chrome browser.

Impact of the GraphicsFuzz tool. The original GLFuzz tool was highly successful in applying metamorphic testing to finding bugs in GLSL compilers from all seven major GPU providers [R5, R6]. Donaldson disseminated this to a wide audience by writing a series of blog posts describing these findings ( AMD post, Apple post, ARM post, Imagination Technologies post, Intel post, NVIDIA post, Qualcomm post). These bugs included three serious security vulnerabilities for which the Imperial team received public recognition – an Apple iOS information leak [I2], an NVIDIA kernel mode vulnerability [I3], and data-capture exploit caused by a defect in an ARM graphics driver [I4]. Donaldson and team promoted GraphicsFuzz on social media, reaching the front page of HackerNews and being featured by sites such as Phoronix [I5].

Acquisition of GraphicsFuzz Ltd. by Google, and open sourcing. Encouraged by this media interest, Donaldson and his post-docs Hugues Evrard and Paul Thomson secured funding from the DCMS/Innovate UK ICURe programme to investigate commercialisation opportunities for the work, leading to them founding GraphicsFuzz Ltd., incorporated in December 2017 and spun-out from Imperial in February 2018 [I6]. The team promoted the technology via an innovative web-app that would run a set of GraphicsFuzz tests on a user’s device, allowing them to tweet a summary of the test results [I7]. This raised the profile of GraphicsFuzz quickly, leading to the acquisition of the company by Google in August 2018 for an undisclosed sum [I6]. Google open-sourced the GraphicsFuzz tool in September 2018 [I8].

Ongoing impact of GraphicsFuzz on graphics driver quality. Donaldson, Evrard and Thomson joined Google UK, and now lead a team specialising in methods to improve the quality of Android graphics device drivers, deployed on billions of devices worldwide. GraphicsFuzz is being used to find defects in Android graphics drivers from ARM, Qualcomm, Imagination Technologies and NVIDIA, who collectively produce GPUs for more than 79% of the smartphone and tablet GPU market. Tests that expose these bugs are being integrated with the Android Compatibility Test Suite, meaning that devices can only ship future versions of Android if these bugs are fixed in their drivers. The impact of this ongoing use of GraphicsFuzz is detailed in the supporting letter from Google’s Director of Engineering, Android Gaming and Graphics [I1].

Impact on Chrome. The WebGL technology brings graphics rendering capabilities to web applications but exposes widely-used browsers such as Chrome – used by most of the world’s internet users – to graphics-related security vulnerabilities. GraphicsFuzz has been integrated with the ClusterFuzz project for continuous fuzz testing of Chrome, and since January 2019 has found more than a dozen security vulnerabilities in Chrome’s WebGL implementation that had been missed by other testing methods. These vulnerabilities have now been fixed [I9].

Impact on open source graphics drivers. Google has used GraphicsFuzz to rigorously test AMD’s open-source driver for Vulkan, finding a large number of driver bugs, which have been now fixed, including bugs that turned out to affect the widely-used LLVM compiler framework [I10]. GraphicsFuzz has also been used to find defects in the open-source Mesa drivers for Intel and NVIDIA GPUs [I11].

Tools for the Vulkan ecosystem. GraphicsFuzz has found more than 100 bugs collectively in a suite of tools that underpin the Vulkan graphics programming model: the glslang shader translator, the spirv-opt and spirv-val optimizer and validator that ship as part of the SPIRV-Tools framework, and the spirv-cross cross-compiler [I12].

Integration with Vulkan CTS. Every Vulkan driver must pass the Vulkan Conformance Test Suite (CTS), which now features a new graphicsfuzz test group, comprising tests generated by GraphicsFuzz that exposed bugs in previously-conformant Vulkan drivers. Adding these tests to the Vulkan CTS requires fixes to any defective drivers in order for them to remain conformant. The Vulkan CTS now features more than 300 GraphicsFuzz tests [I13].

Impact of GPUVerify on ARM. The collaboration between Imperial College London and ARM during the CARP EU project (coordinated by Donaldson) led to ARM integrating GPUVerify into their Mali Graphics Debugger [I14]. Leveraging fundamental results on how to scale analysis to thousands of parallel threads [R1, R2], this allows software developers targeting ARM’s Mali GPU series – which leads the smartphone and tablet GPU market – to reason rigorously about the OpenCL programs that they write.

Impact of CLsmith on OpenCL compilers. The Imperial team who developed CLsmith initially used it to find more than 50 bugs in OpenCL compilers from companies including Altera, AMD, ARM, NVIDIA and Intel [R3, R4]. CLsmith has been used by Codeplay Software Ltd., a UK-based compiler company, for continuous integration of their OpenCL compilers; see letter of support from Codeplay’s Principal Software Engineer, AI Parallelism [I15]. Via a TETRACOM technology transfer project, CLsmith has since been incorporated into the set of tools used by UK-based start-up company dividiti for analysis of many-core platforms [I16].

Beneficiaries of the impact: The immediate beneficiaries of GPUVerify, CLsmith and GraphicsFuzz are GPU technology developers, who are now able to construct more reliable software and drivers as a result of the impact of these tools. End users of computer systems (i.e. the general public) are ultimately beneficiaries due to more reliable applications that use computer graphics.

Nature of the impact: By finding bugs in GPU technology automatically, these tools (a) reduce the manual effort required for GPU quality assurance, (b) increase test rigour by finding defects that human testers would be unlikely to uncover, and (c) increase the coverage of graphics driver compatibility test suites.

The impact arising from this research during the REF period of 1 August 2013 and 31 December 2020 is related to Facebook acquiring Infer, a start-up company and software tool based on the fundamental research described above, and the wide industrial adoption of the Infer tool. Facebook has integrated the Infer tool with their continuous development workflow, discovering and fixing a large number of bugs in products used by billions of users. After making the technology available as open source, there has been broad adoption by companies across the whole software industry.

Background to the Facebook acquisition. Based on the success of the foundational research described above for enabling scalable reasoning about pointer-manipulating programs, Calcagno and Distefano (based at Imperial College London and Queen Mary University of London, respectively) incorporated a start-up company, Monoidics Ltd. (company no. 06805518), in January 2009, as directors. They developed the Infer tool, an implementation of the line of research (papers R1 – R6 above) that culminated in the Journal of the ACM article [R6]. Monoidics was funded by Quantum Wave Capital, received support from the CARP EU FP7 project (led by Imperial College London), and reported ARM, Airbus and Mitsubishi Electric as customers. In July 2013, Facebook acquired the assets of Monoidics for an undisclosed sum and hired the engineering team, including Calcagno [I1]. Calcagno has since worked as a Software Engineer at Facebook, remaining a Lecturer on Leave of Absence at Imperial until 2017, and now retains his connection with Imperial as an Honorary Lecturer.

Impact of Infer at Facebook. Facebook integrated Infer into its perpetual software development process [I2]. As a result, Infer is now used to continuously analyse the multi-million-line code bases of both mobile apps and back-end services at Facebook, as well as the software associated with WhatsApp and Instagram [I3]. Each time a developer commits any code change, Infer analyses the change and participates in the code review process, indicating potential memory safety problems that the programmer may have overlooked. This is possible due to the scalability afforded by bi-abduction [R6]: only the procedures related to the code change must be re-analysed, meaning that analysis can complete within minutes, providing a report of potential problems while the code changes are fresh in the mind of the developer.

At a talk at Imperial College London in 2015, Facebook CTO Mike Schroepfer said about Infer: “It’s a really key component of being able to build an app and ship it weekly to a billion people. If you want to do this without everything grinding to a halt, it’s really helpful to have a system that automatically detects and helps with bugs” [I4]. In a 2019 Communications of the ACM article, Facebook explains that Infer now runs continuously on millions of lines of code, and that so far more than 100,000 bugs found by Infer have been fixed before affecting end users. It reports a high fix rate of 70% associated with memory safety bugs flagged by Infer [I3].

The high degree of impact that Infer continues to have, and its underpinning by the original research of Calcagno et al., is summed up well in the letter from Facebook’s Director of Engineering [I5]: “Today within Facebook, the verification technology originally designed by Prof Dino Distefano, Dr Cristiano Calcagno and colleagues contributes to the quality of the products we serve to over 2 billion people every day. Currently Infer analyses every applicable code change within Facebook and its family of apps (Messenger, WhatsApp, Instagram, Oculus). Every month, thousands of issues are fixed before these apps are shipped to users. Thus, Infer has saved Facebook engineers innumerable hours that they would have had to spend debugging the problems it detected, had those problems reached production.”

The significance of the link between the fundamental research and its impact was recognised by the Infer team receiving the prestigious 2016 CAV Award, which “recognizes their contribution to the development of Separation Logic, the theory that underpins the Infer static analyser”.

Impact of Infer beyond Facebook. Infer was made open source in 2015 [I6], making it available to other companies. The open source repository is very active (>150 contributors) and popular (starred by >12,000 GitHub users, forked >1,600 times). Companies who have publicly acknowledged using Infer include Amazon Web Services, Mozilla, Oculus, Spotify, Uber, Freefem, JD.com, Marks and Spencer, Sky, Money Lover, LA, Netcetera, Vuo, CodeAI, and Tile [I7]. The National Cyber Security Centre (NCSC, part of GCHQ) also uses Infer. We detail here the four most significant uses of Infer in industry and government beyond Facebook: by Spotify, Uber, Muse Dev (a recent start-up company), and NCSC.

Spotify: The Android Infrastructure team at Spotify have extended Infer in order to incorporate it in their continuous integration process for Android. In an article describing the collaboration [I8], A Spotify engineer writes: “In our quest to make our code even better, we started using Infer. Infer found several legitimate issues that other tools had missed. The Infer team was also very helpful in following a few false positives that we encountered, and we now have it running on our build servers.”

Uber: Uber have reported using Infer in a portfolio of tools for static detection of null pointer exceptions, claiming that this endeavour “reduced the number of NPEs observed in our apps in production by an order of magnitude” [I9].

MuseDev: A spin-off company from Galois, MuseDev is a new company providing tools and services for automated code reasoning. As detailed further in the supporting letter from MuseDev CEO [I10], MuseDev runs a cloud-based code analysis service that uses Infer as its flagship analyser: “Via Muse, Infer is now running daily in the cloud on several open source repositories. And we are actively working with a variety of enterprises to deploy and run MuseDev in their environments, including Fortune 500 and fast-growing venture-backed technology companies.” The “cool math” link on the MuseDev website is to a review article on separation logic, with a section entitled “Bi-Abduction and Facebook Infer” that indicates the importance of the underpinning research. Magill further stresses the breakthrough nature of this research: “The technology developed at Imperial and Queen Mary ushered in a new wave of static analysis, capable of finding subtle and important bugs while running quickly enough to fit into modern agile development workflows. In fact, in talking with customers, we often use the terms ‘first wave’ and ‘second wave’ to distinguish just how different these techniques are from what came before.”

National Cyber Security Centre (NCSC): This case study is accompanied by a confidential letter from the Head of High Assurance Engineering Research at NCSC (available to reviewers of the case study), outlining the use of Infer by NCSC and the significance of the fundamental research on which Infer is based [I11].

Beneficiaries of the impact: The companies that use Infer are direct beneficiaries of the impact: 18 large organisations publicly advertise that they use the tool [I7], and the open source. The improvement in software quality afforded by Infer is to the benefit of the users of the products that these companies build. Facebook alone report that Infer has led to more than 100,000 fixed bugs in the Facebook, WhatsApp and Instagram code bases [I3].

Nature of the impact: Software bugs are expensive: a report from the Consortium for IT Software Quality (CISQ) estimates that the cost of poor-quality software in the US alone is USD2.84 trillion. By finding bugs early during the software development process, Infer helps reduce these costs, leading to major economic impact. Billions of people worldwide depend on products developed by the companies that use Infer. The large number of software defects in these products that have been fixed during development, after identification by Infer, and that would otherwise have reached production to the detriment of end users is evidence of the broad societal nature of the impact [I3].

The underpinning research has led to the development of novel imaging biomarkers that are now routinely used in clinical trials to assess the efficacy of new drugs and treatments. The biomarkers are also used in healthcare diagnostics, e.g., for dementias such as Alzheimer’s disease (AD).

Economic impact

To maximise the economic impact of the research, a team of Imperial researchers (Rueckert, Hajnal) started IXICO with colleagues from UCL (Hawkes, Hill) in late 2004. In the assessment period for the 2014 REF, IXICO had been established and grown organically. In the current REF assessment period, several new key impacts occurred: in particular, in October 2013, it became IXICO plc and is now listed on the Alternative Investment Market (AIM) of the London Stock Exchange (date: 15th October 2013). Other impact measures are listed in the table below and compared to the status at the beginning of the assessment period:

Since the start of the assessment period, IXICO has grown from 36 to 78 employees. IXICO’s revenues have increased from GBP2,500,000 to more than GBP9,500,000. IXICO has become a profitable business and, since 2014, has won more than GBP36,200,000 in business from the global pharmaceutical industry (GSK, Pfizer, Bristol-Myers Squibb, Novartis, Eli Lily). Its orderbook for 2020 is GBP21,700,000 alone. IXICO’s image analysis technology is based on the underpinning research described in Section 2 and has been transferred from Imperial during IXICO’s formation and later as part of an IP pipeline agreement with Imperial. It has been, and/or is currently being used to analyse tens of thousands of medical images collected from a total of more than 2,000 imaging centres across North America, Latin America, Europe, Asia and Australasia (in over 50 countries) [I1, I2, I3]. Since 2013, IXICO has been involved in over 50 clinical trials and analysed brain images from more than 100,000 subject visits using its image analysis technology [I2].

Originally, IXICO had focused on applying its image analysis technology to supporting clinical trials of novel therapies to treat Alzheimer’s disease. Since the beginning of the REF assessment period, IXICO has successfully translated the technology to other neurological diseases, including Huntington’s disease, Parkinsonian diseases (such as PD, PSP and MSA), Multiple Sclerosis and rare neurological diseases. In particular, IXICO has had much success in the Huntington’s disease (HD) clinical trials market, having built the requisite therapeutic expertise and validated the technology specifically for measuring imaging biomarkers that are correlated with disease pathology and progression in HD. Since 2014, IXICO has been selected for 10 clinical trials in HD, 7 of which were started over the past 4 years since 2016 and includes the world’s first phase III trial in Huntington’s disease for a gene therapy approach that targets the mutant Huntington gene [I1].

The developed image analysis techniques also provide an important tool in clinical trials to assess the safety and efficacy of new drugs as well as for enrichment of clinical trials, i.e., as inclusion/exclusion criteria. This enrichment of trials enables the reduction of heterogeneity of the population providing additional power for clinical trials, therefore increasing the sensitivity of trials to detect disease-modifying effects of drugs. Since 2013, IXICO has used the automated image analysis techniques for safety and eligibility reports in over 20,000 cases [I2].

Impact on public policy and services

The techniques for automatically computing imaging biomarkers such as low hippocampal volume have had a significant impact on informing the development of new guidelines in regulatory clinical trials. This has been recognised, e.g., by the Coalition Against Major Diseases (CAMD) consortium by submitting to regulators an application to qualify low hippocampal volume as a biomarker [I4, I5]. This submission – supported by the US Federal Drug Administration – incorporates key data obtained using the underpinning research described here: the availability of this technology, with the regulatory qualification, is having a global impact on the design of future trials of AD medicines in the pre-dementia population. Furthermore, IXICO is also part of key public/private partnerships that are advancing the scientific and regulatory agenda for clinical trials in neurological diseases. These include:

• EPAD (the European Prevention of Alzheimer’s Dementia; http://ep-ad.org) – which has built an adaptive clinical trials platform with a readiness cohort of people at risk of Alzheimer’s disease and which will enable pharma companies to evaluate efficiently the safety and effectiveness of their medicines in patients prior to onset of symptoms by utilising adaptive clinical trial approaches. IXICO received over EUR600,000 in grant funding to provide the technology to collect, manage and analyse MRI scans.

• AMYPAD (Amyloid Imaging for Prevention of Alzheimer’s Dementia, https://amypad.eu) – which is evaluating the diagnostic value of amyloid imaging in the diagnosis of Alzheimer’s disease. IXICO received over EUR1,000,000 in grant funding to provide the technology to collect, manage and analyse MRI and PET scans.

• HD-RSC (Huntington’s Disease Regulatory Science Consortium, https://c-path.org/programs/hdrsc/) – which is developing regulatory-endorsed drug development tools to support the development and approval of new medicines to treat HD and involves representatives from the FDA. IXICO is contributing to the regulatory clearance of imaging biomarkers.

• CPP (Critical Path for Parkinson’s, https://c-path.org/programs/cpp/) – which is developing regulatory-endorsed drug development tools to support the development and approval of new medicines to treat Parkinson’s disease and involves representatives from the FDA. IXICO is contributing to the development of a regulatory-endorsed digital outcome measure for Parkinson’s disease.

Beneficiaries of the impact

The beneficiaries are pharmaceutical companies developing new drugs and therapies for neurodegenerative diseases such as Alzheimer’s or Huntington’s disease through improved efficacy and safety of clinical trials. Furthermore, patients are benefitting from improved imaging-based diagnostics and therapeutics.

Nature of the impact

IXICO’s image analysis techniques provide an important tool in clinical trials to assess the safety and efficacy of new drugs as well as for enrichment of trials, enabling the reduction of heterogeneity of the population. The image analysis techniques provide increased sensitivity to detect disease-modifying effects of drugs and trials. This means that fewer patients need to be recruited into trials, leading to significant cost savings for pharma companies. IXICO has also supported clinical trials of novel therapies relating to neurological diseases and contributed to new guidelines for biomarkers in regulated clinical trials.

Impact on robotics products and R&D investment

Dyson product development: Davison worked as an academic consultant to Dyson until 2014 on the long-term project of developing their first robot product (Dyson 360 Eye), which makes direct use of the published algorithms in [R1, R2], and incorporates them into Dyson’s robotic vacuum cleaner development. He was responsible for guiding Dyson engineers to successfully design an omnidirectional visual SLAM system, which was the unique key feature (and thus chosen as the name) of the Dyson 360 Eye released worldwide in 2015. Dyson is one of few large UK companies to have a leading robotic product already on the market.

In 2018, Dyson’s second robotic product, the Dyson 360 Heurist [I2], which also incorporates Davison's core SLAM system, went on sale worldwide. The Principal Robotics Architect at Dyson states “ Andrew's pioneering research on vision-based SLAM in published algorithms such as MonoSLAM and Inverse Depth Features directly inspired the visual mapping system Dyson developed for the robot, based on a 360-degree panoramic camera. This enabled the robot to navigate precisely and repeatedly, and therefore clean rooms in a much more direct and efficient manner than other products on the market […] The core visual SLAM system developed for the 360 Eye with Andrew's input lives on in Dyson's latest and current robot vacuum cleaner, the Dyson 360 Heurist, which is now on sale worldwide” [I1].

In addition, Imperial’s SLAM research led to Dyson making large ongoing investments in robotics as a core focus of their business. At the start of Davison’s consultancy, the internal Dyson robotics team had fewer than 5 people, and now Dyson employs hundreds of scientists and engineers in robotics R&D in both the UK and Singapore [I1]. In 2020, Dyson announced a new investment of GBP3 billion specifically into AI, robotics and batteries [I9]. Since 2014, the Imperial team continues to collaborate closely with Dyson, who funded the Dyson Robotics Laboratory at Imperial directed by Davison and Leutenegger [I3]. The lab has produced over 10 patents for technologies selected by Dyson as relevant to their future robotics product lines.

Impact on augmented reality products and R&D investment

The Imperial research on sparse, dense and semantic SLAM has also directly impacted the augmented reality industry, specifically VR and AR products in headsets and smartphones, which need accurate but low-cost tracking and scene reconstruction from their on-board cameras. Major multi-billion dollar industry projects such as Facebook/Oculus’ VR/AR platforms, Microsoft Hololens, and Google’s ARCore have acknowledged that their algorithms used for sparse and dense mapping are derived from Imperial research work.

Google initiated Project Tango in 2013 to put SLAM into smartphones. In 2018, this technology was transitioned into ARCore, a major platform which enables developers to build AR applications for Android, now used in over 250 million devices worldwide, for functionality such as the AR walking directions feature of Google Mobile Maps. The Engineering Director at Google states [I11]: “ *Imperial College’s visual SLAM research led by Andrew Davison and Stefan Leutenegger was one of the initial inspirations behind Project Tango, going back to the first demonstration of real-time single camera MonoSLAM which showed the feasibility of vision-driven localisation and mapping using low-cost sensors. […] As the market for spatial aware computing evolves, research from Imperial College on SLAM and semantic understanding of the environment will continue to grow in demand. In my opinion, the impact of the technology will be similar to that of the Global Positioning System (GPS) which has transformed many existing industries, and made new ones possible. Wearables, smart home devices, vehicles, and robotics will all depend heavily on a shared understanding of physical space.*”

Surreal Vision acquisition: In 2013, Imperial PhD students (Newcombe, Lovegrove and Salas-Moreno) co-founded the AR company Surreal Vision, advised by Davison and Kelly, to commercialise the dense and object-based SLAM research in DTAM [R4] and SLAM++ [R5]. Based on this work, Surreal Vision developed a new level of accurate but efficient 3D reconstruction algorithms for VR and AR systems, able to provide a rich model of an environment including people and their interactions, but still running on modest consumer computing hardware. Surreal Vision was acquired for USD40,000,000 by Oculus (part of Facebook) in 2015 [I4]. The Surreal Vision team is now part of Facebook Reality Labs, a world leading industrial research team in AR for telepresence. Newcombe is Director of Research Science and leads a team of over 100 researchers which created the fundamental LiveMaps SLAM technology that originates from Imperial’s research. This is a key part of Facebook’s future technology plans for AR as a cloud platform for building, sharing and annotating always up-to-date scene maps across a multitude of user devices [I8].

Microsoft product development: Imperial’s SLAM work was incorporated in Microsoft’s Hololens AR headset product line through multiple stages. MonoSLAM [R1] directly impacted the development of Hololens’ core sparse visual tracking system [I5]. After that, in 2010, KinectFusion [R3], a collaboration between the Imperial team and Microsoft Research Cambridge, became a key Microsoft technology and is now part of Microsoft’s official Kinect for Windows SDK, which is used in millions of Kinect Sensors [I6]. KinectFusion also enables the dense mapping capabilities of Microsoft’s Hololens 1 (on sale in 2016) and Hololens 2 (released in 2019 and now on sale) products, which feature Kinect-like depth sensors [I5].

The Partner Director of Research, Microsoft Research Cambridge states “ [t]he SLAM research at Imperial College led by Andrew Davison has impacted the development of HoloLens significantly. Andrew's research on real-time MonoSLAM was an early inspiration in the development of the core tracking system in HoloLens. HoloLens used a sparse feature mapping visual SLAM approach to achieve the robust and accurate inside-out tracking of the wearer’s head which is crucial for a convincing user experience of mixed reality. […] KinectFusion was very important in impacting the approach we currently use for environment mapping in HoloLens 1 and 2. KinectFusion pioneered real-time volumetric fusion of depth data, and the way that this can be implemented efficiently using parallel computation. […] the KinectFusion algorithm for dense environment mapping based on measurements from a moving depth camera […], originally targeted Microsoft's newly release Kinect device, and was hugely influential guiding research in scene mapping for multiple applications including robotics, with the unprecedented detail and speed of reconstruction it enabled. […] On the HoloLens, a derivative of the KinectFusion algorithm allows a real-time reconstruction of the environment around the user, allowing virtual objects to be aligned with real world objects and occlusions of the virtual by the real to be handled appropriately” [I5, I6].

Impact on software development in robotics industry

SLAMcore is an Imperial spin-out company, co-founded in 2016 by Leutenegger, Davison, two PhD students (Zienkiewicz and Kim) and a lab manager (Nicholson). As of 2020, it has over 20 employees and received investment from the UK, US and Japan of over GBP7,000,000 to develop Spatial AI solutions for robots and drones [I7].

SLAMcore has impacted the market of spatial robot and drone software development by offering a licensable software development kit (SDK). The SLAMcore Spatial AI Software Stack SDK enables robots to determine their position and map their surroundings in detail to enable long-term intelligent operation from on-board cameras and other sensors. SLAMcore has established formal hardware partnerships with Intel, NVidia and ARM [10]. The SDK currently is licensed by monthly subscription by 8 customer companies in applications such as warehouse management and evaluated by over 50 more that require SLAM technology for new robot products.

Beneficiaries of the impact

Imperial’s vision-based SLAM research has led to key low-cost localisation and mapping algorithms used by major international technology companies, such as Dyson, Facebook, Google and Microsoft, enabling development of new spatially intelligent consumer products and services. Understanding their context in the surrounding environment enables products such as autonomous cleaning robots to relieve users from household cleaning chores and human-centred, mixed-reality headsets to facilitate new forms of entertainment as well as workplace tasks.

Nature of the impact

Vision-based SLAM is an enabling technology for low cost, high performance, mobile robot products emerging in the consumer sector. The research has also directly impacted development and products in the growing area of mixed-reality devices, which global technology companies now regard as a major new wave in future human-computer interaction.