Â鶹´«Ã½

• Inactive for Year: 2026/27

• Module Leader(s): Professor Wei Pan
• Lecturer: Professor Damian Giaouris

• Owning School: Engineering
• Teaching Location: Â鶹´«Ã½ City Campus

Semesters

Your programme is made up of credits, the total differs on programme to programme.

Aims

This module is concerned with autonomous systems that integrate AI components — learned models, policies and controllers — and must operate reliably in physical environments with real constraints. Building on foundations in estimation, modelling and control from Semester 1, and running in parallel with the Reinforcement Learning module, the focus is on the system-level challenges that arise when moving from simulation to hardware: how do you ensure that a system built around a learned component works reliably, safely and robustly on a real platform?

The module covers the practical and theoretical issues faced by such systems, including uncertainty, limited sensing, energy constraints, communication delays, actuation limits, and safety-critical consequences of failure. Topics include sim-to-real transfer, robustness of learned policies within the wider system, online adaptation when operating conditions change, runtime safety monitoring, and multi-robot coordination.

Consideration is also given to the regulatory frameworks and ethical implications associated with deploying autonomous systems that rely on AI, which are increasingly important in industry and public policy.

Outline Of Syllabus

From simulation to real-world platforms: the sim-to-real gap, simulation fidelity, domain randomisation, domain adaptation, transfer of learned models and policies to physical platforms

Combining learned and classical controllers: model-reference control with learned components, when to trust a learned policy and when to fall back to a classical controller, handling actuation limits, energy and latency constraints

Online adaptation: model learning under non-stationarity, continual learning, detecting and responding to out-of-distribution inputs, knowing when your model is wrong

Robustness: adversarial perturbations, domain randomisation for robustness, worst-case performance analysis, robust policy evaluation

Safe deployment: runtime monitoring, safe exploration strategies, fallback architectures, verification and validation of learned components, safety envelopes and constraint enforcement, formal safety constraints, barrier functions and Lyapunov-based stability analysis

Multi-agent deployment: consensus and coordination, distributed estimation, decentralised decision-making with communication constraints, cooperative and competitive multi-robot scenarios, formation control, multi-agent reinforcement learning

Regulation, ethics and societal impact: current standards for autonomous systems, accountability, transparency, public trust

Case studies from mobile robotics, autonomous vehicles, multi-robot systems and industrial applications

Teaching Methods

Teaching Activities

Teaching Rationale And Relationship

Lectures follow the deployment pipeline: sim-to-real first, then robustness, then safety, then multi-agent systems, then regulation (Knowledge Outcomes 1-6). This mirrors the order in which these issues arise in practice when moving a system from the lab to the field.

Practical sessions are where students get hands-on with the deployment challenges discussed in lectures (Skill Outcomes 1-3, 5). Practicals focus on individual techniques (e.g. applying domain randomisation, setting up a runtime monitor).

Guided reading each week covers a mix of research papers and industry reports. This is a fast-moving area, so the reading list will be updated regularly. Students are expected to engage critically with the material, not just summarise it (Skill Outcome 4).

Independent study time supports exam revision and wider reading.

Assessment Methods

The format of resits will be determined by the Board of Examiners

Exams

Formative Assessments

Formative Assessment is an assessment which develops your skills in being assessed, allows for you to receive feedback, and prepares you for being assessed. However, it does not count to your final mark.

Assessment Rationale And Relationship

The exam tests whether students understand the theory behind sim-to-real transfer, robustness, safety and multi-agent deployment (Knowledge Outcomes 1-6). It includes scenario-based questions where students have to diagnose deployment problems and propose solutions, as well as more standard analytical questions.

Formative exercises are provided throughout the semester, including practical checkpoints that give feedback on the techniques before they are used in the group project.

Alternative forms of assessment are available where required. These could include oral exams and digital exams

Reading Lists

Timetable

• Timetable Website: