Diffusion-based Deep Generative Models for Assessing Safety in Autonomous Vehicles.

It is hoped that the use of autonomous vehicles will significantly reduce the number of road accidents due to human error. However, extensive testing will be necessary to demonstrate that they satisfy a high standard of safety before they can be introduced. Much of this testing must be carried out in simulated environments, which allow for a far greater degree of safety and flexibility than road testing. This is done by testing the vehicle AI on a set of predetermined scenarios. Examples of scenarios include recovering from a loss of control due to road conditions, performing manoeuvres in the presence of oncoming traffic, and responding to a sudden deceleration by a leading vehicle. A key issue in autonomous vehicle safety is the need for a huge, diverse set of scenarios that reflect both normal driving conditions and difficult situations. This project aims to develop methods for automating scenario generation, both by reconstructing conditions from real life driving datasets, and by creating completely new scenarios using statistical generative modelling techniques from the statistical AI literature, such as generative adversarial networks and variational auto encoders.
The problem with current generative modelling techniques is to generate realistic scenarios from the same underlying distribution as a given dataset. Diffusion-based models are a recent advance in generative modelling, which use stochastic differential equations to gradually transform random noise into data via a diffusion process. This is learned by training a neural network to remove small amounts of noise from existing data. They are efficient to train and have been shown to be effective at generating diverse samples from complex, high-dimensional distributions. This research project will focus on the statistical properties of diffusion models and how they can be adapted for the scenario generation problem to generate realistic new scenarios given an existing database.

In partnership with Transport Research Laboratory.

This proposal will benefit (i) the UK economy and society, (ii) our industrial partners, (iii) the wider community of non-academic employers of doctoral graduates in STOR, (iv) the scientific disciplines of statistics and operational research and associated academic communities, (v) UK doctoral students in STOR, and (vi) the CDT students themselves.

Below we outline how each of these communities will realise these benefits:

(i) The UK economy will gain a competitive edge through a significant increase in the supply and diversity of doctoral STOR professionals with the skills required to undertake influential, responsible and impactful research, and who have been trained to become future leaders. Our goal is that our future alumni who enter industry assume leading roles in realising the major impact that STOR can make in achieving effective data driven decision-making. Our existing alumni are already starting to achieve this. A wider societal benefit will accrue from research contributions to EPSRC Prosperity Outcomes, e.g. to the UK being a Productive and Resilient Nation.

(ii) Our industrial partners will particularly benefit from the skills supply identified in item (i), as likely employers of STOR-i graduates. They will further benefit from teaming with a community of leading edge STOR researchers in the solution of substantive industrial challenges. Mechanisms for the latter include doctoral projects co-supervised with industry, industrial internships, engagement in research clusters and industrial problem-solving days. Our training programme will give students the skills they need to ensure that research is conducted responsibly and that outcomes are successfully communicated to beneficiaries. The value that our industrial partners place on working with STOR-i can be seen through the pledged cash support of £1.7M.

(iii) A wider benefit will accrue from the employment of STOR-i graduates, equipped as described in items (i) and (ii), across non-partner public and private sector organisations. The breadth and depth of training provided by the CDT will enable students to quickly make a difference in these organisations, using their research skills to affect significant change.

(iv) The STOR academic community will benefit from methodological advances and from the increase and diversity in the supply of STOR researchers who value, and have experience of, collaborative research. Our alumni will be leaders in 21st Century Statistics with a strong culture of, and training in, reproducible research and a focus on achieving impact with excellence. Our recruitment strategy will further benefit this community in achieving a healthier supply of high-quality doctoral candidates from diverse backgrounds. Our research internship programme gives top mathematically able individuals from across the UK an experience of STOR research and has been shown to increase applications for STOR PhD programmes across the UK.

(v) Elements of the STOR-i programme will benefit the wider community of UK doctoral students in STOR. Using financial support from our industrial partners, we will continue our National Associate Scheme. This will provide up to 50 UK STOR doctoral students with funding and access to elements of STOR-i's training programme. An annual conference will provide opportunities for learning, networking and sharing research progress to members of the scheme.

(vi) STOR-i students will benefit from a personalised programme that will support each individual in fully achieving their research leadership potential, whether in academia or industry. Students will be given the tools and opportunities to develop research and broader skills that will enable them to achieve maximum scientific impact for their work. Our current alumni provide strong evidence that these future graduates will be extremely employable.