Master'sData, AI, Simulation & Analytics Applications open

Master’s in Reinforcement & Imitation Learning for Control

Online master's programme of 1,900 hours in reinforcement and imitation learning for control: RL/IL, hybrid MPC and safety. 12 modules covering sequential decision-making, tabular and deep RL, policy gradients, model-based RL, imitation and offline learning, sim-to-real transfer and safe deployment, with ISO 13849-1, IEC 61508 and ISO 10218-1, and an assigned technical tutor.

Duration12–14 months

ModeOnline

LanguageEnglish

Start datesSet by the faculty

AwardSEIUM Institute diploma

Enrol now View syllabus Ask an advisor

Master’s in Reinforcement & Imitation Learning for Control — studying from Singapore

Singapore has invested heavily in artificial intelligence and data, under national plans coordinated by the Infocomm Media Development Authority and the AI Singapore programme. The sector spans industrial analytics, digital twins, simulation for engineering design and data platforms serving finance, logistics and healthcare. Engineers work across machine learning, data engineering and computational modelling. Employers range from global technology firms and banks to local research institutes and government-linked companies.

Fees for Singapore: S$9,880 (Singapore dollar), with instalments and employer sponsorship available.

Relevant bodies in Singapore

Infocomm Media Development Authority (IMDA)
AI Singapore
Personal Data Protection Commission (PDPC)
Singapore Computer Society

Typical roles

Machine Learning Engineer
Data Engineer
Simulation Engineer
Analytics Consultant

Questions from Singapore

Does Singapore have a recognised body for data and AI professionals, and is SEIUM recognised?

Singapore's Infocomm Media Development Authority (IMDA) and the Singapore Computer Society are key bodies in the digital and data field, and the Personal Data Protection Commission (PDPC) oversees data protection. SEIUM's qualifications are private, not state degrees, and not accredited by any national body. Acceptance is decided by employers or the relevant body, so verify this before enrolling.

Which Singapore employers hire data, AI and simulation specialists?

Demand is strong across banks, government agencies, healthcare, logistics, telecoms and tech firms. Singapore's national AI and smart nation initiatives also create roles. Employers typically assess portfolios, technical interviews and proven projects. Your final project can demonstrate applied analytics or simulation skills, though hiring decisions are the employer's.

Can I study this part-time from Singapore, and does it need a student visa?

Yes, the programmes are 100% online and taught in English, so you can study from Singapore without a student visa. Workload is substantial, with 1,900 hours for a master's, 800 for a postgraduate diploma and 300 for a specialist course. Fees can be paid in instalments, and employer sponsorship is possible, subject to your employer's approval.

Virtual lab · try it before you enrol

Take the controls.

Run a real engineering trade-off from this field. Move the sliders, watch the model react in real time and beat the challenge.

SIM · LIVE ●

Learning outcomes

What you will be able to do

Formulate control problems as sequential decision processes.

Implement tabular and deep reinforcement learning algorithms.

Apply actor-critic methods to systems with continuous actions.

Learn policies from expert demonstrations.

Transfer policies from simulation to the real system.

Incorporate safety constraints into learned control.

Deploy learned controllers in industrial environments.

Profile

Who it's for

Reinforcement learning engineers who want to take policies from simulation to real systems.
Advanced control engineers looking to combine MPC with learned policies.
Robotics engineers working on manipulation, locomotion or autonomous vehicles.
AI researchers focused on control and decision-making.
Intelligent automation engineers in process industries and buildings.
Robotics simulation specialists who build training environments.
Engineering or science graduates looking to enter robotics.

Standards & tools

What you'll work with

ISO 13849-1 (safety of machinery — safety-related parts of control systems)
IEC 61508 (functional safety of electrical/electronic/programmable electronic safety-related systems)
ISO 10218-1 (safety requirements for industrial robots)
Regulation (EU) 2024/1689 (AI Act)
Regulation (EU) 2023/1230 (Machinery Regulation)
ISO/IEC 22989 (artificial intelligence concepts and terminology)
Gymnasium
Stable-Baselines3
MuJoCo
NVIDIA Isaac Lab
ROS 2

Syllabus

12 modules

1.1From classical control to learned control
1.2Markov decision processes
1.3States, actions, rewards and policies
1.4Bellman equations
1.5Dynamic programming and value iteration
1.6Partial observability and POMDPs
1.7Relationship with optimal control and LQR
1.8Cases in robotics and processes
1.9Limits and risks of learning in control
1.10Case study: MDP formulation of a thermal system

2.1Monte Carlo methods
2.2Temporal differences and TD(λ)
2.3Q-learning and SARSA
2.4Exploration versus exploitation
2.5Multi-armed bandits
2.6Eligibility traces
2.7Convergence and learning rate
2.8Reward design
2.9Environments with Gymnasium
2.10Lab: Q-learning on a discretised pendulum

3.1Function approximation with networks
3.2DQN and the replay buffer
3.3Double DQN and dueling
3.4Prioritised experience replay
3.5Rainbow DQN
3.6Distributional RL
3.7Instability and target networks
3.8Hyperparameter tuning
3.9Reproducible evaluation
3.10Practical exercise: DQN for controlling a discrete system

4.1Policy gradient theorem
4.2REINFORCE and baselines
4.3Actor-critic and A2C
4.4Trust region and TRPO
4.5PPO and objective clipping
4.6DDPG and continuous actions
4.7TD3 and bias reduction
4.8Soft Actor-Critic and entropy
4.9Algorithm comparison
4.10Lab: continuous control of a robotic arm with SAC

5.1Learned dynamics models
5.2Dyna and planning
5.3Model predictive control with learned models
5.4PETS and probabilistic ensembles
5.5Dreamer and world models
5.6MuZero and tree search
5.7Model uncertainty
5.8Sample efficiency
5.9Combining with physical models
5.10Practical exercise: MPC with learned dynamics

6.1Behaviour cloning
6.2Covariate shift
6.3DAgger and data aggregation
6.4Inverse reinforcement learning
6.5GAIL and adversarial learning
6.6Demonstrations via teleoperation
6.7Quality and diversity of demonstrations
6.8Imitation from observations
6.9Diffusion policies
6.10Lab: imitating an expert controller

7.1Reinforcement learning with historical data
7.2The extrapolation problem
7.3Conservative Q-learning
7.4Implicit Q-learning
7.5Offline policy evaluation
7.6D4RL datasets
7.7Industrial operating data
7.8Combining offline and online learning
7.9Validation before deployment
7.10Case study: policy from plant logs

8.1MuJoCo and Isaac Gym simulators
8.2Massively parallel simulation
8.3Domain randomisation
8.4System identification
8.5Online adaptation
8.6Delays and sensor noise
8.7Sim-to-real gap in locomotion
8.8Digital twins for training
8.9Test-bench validation
8.10Practical exercise: training in simulation and testing on hardware

9.1Constrained reinforcement learning (CMDP)
9.2Safety shields
9.3Control barrier functions
9.4Safe exploration
9.5Adversarial robustness
9.6Neural network verification
9.7Runtime monitoring
9.8Fallback to classical controllers
9.9Safety arguments
9.10Workshop: designing a safety shield

10.1Chemical process control
10.2HVAC and energy efficiency
10.3Grid and storage management
10.4Robotic manipulation
10.5Autonomous vehicles and drones
10.6Logistics planning
10.7Traffic signal control
10.8Published cases and lessons learned
10.9Feasibility criteria
10.10Case study: building HVAC control with RL

11.1Policies on embedded systems
11.2Quantisation and inference latency
11.3Integration with PLCs and ROS 2
11.4Performance monitoring
11.5Safe retraining
11.6AI regulation and liability
11.7Functional safety and learned control
11.8Documentation and traceability
11.9Managing RL projects
11.10Lab: deploying a policy on ROS 2

12.1Defining the control problem
12.2Simulation environment
12.3Classical control baseline
12.4Algorithm selection and training
12.5Imitation or offline data
12.6Robustness and safety
12.7Validation on real hardware or test bench
12.8Comparative analysis
12.9Written thesis report
12.10Master's Thesis: defence before a panel

Methodology, software & labs

How you'll learn

100% online training on the SEIUM campus, with downloadable material, recorded classes and an assigned technical tutor. There are no fixed timetables: you progress at your own pace within each intake, so the programme fits around your job.

Each module combines theory, guided reading of the original standard and applied exercises. The working software is Gymnasium, Stable-Baselines3, MuJoCo, NVIDIA Isaac Lab and ROS 2, in free, academic or evaluation versions. Testing is studied through real procedures: training in simulation with randomisation, policy validation on a test bench and comparison against PID and MPC controllers. Assessment is continuous for each module, plus the final Master's Thesis.

Standards covered: ISO 13849-1, IEC 61508, ISO 10218-1, Regulation (EU) 2024/1689 (AI Act), Regulation (EU) 2023/1230 (Machinery), ISO/IEC 22989.

Assigned technical tutor:every module ends with a deliverable that your tutor marks individually.

Standards-based learning:guided reading of ISO 13849-1, IEC 61508 and ISO 10218-1 applied to real cases, not as a theoretical annex.

Industry software:Gymnasium, Stable-Baselines3, MuJoCo, NVIDIA Isaac Lab and ROS 2, in free, academic or evaluation versions.

Master's Thesis with defence:A learned controller for a real control problem, trained in simulation, made safe and validated on hardware or a test bench against a classical baseline, defended before a panel.

Flexible format:100% online, no fixed timetables, compatible with full-time work.

Capstone project

Build it end to end

Your final project follows the real engineering workflow, step by step, reviewed by your tutor.

MDP formulation

Formulate the chosen control problem as an MDP and build its simulation environment.

Classical baseline

Implement a PID or MPC baseline to benchmark the learned controllers.

Policy training

Train a continuous-control policy with SAC or PPO, optionally bootstrapped from demonstrations.

Safety shield and sim-to-real

Add a safety shield and transfer the policy from simulation to hardware.

Master's Thesis

A safe learned controller validated against a classical baseline, defended before a panel.

Career opportunities

Where this takes you

Reinforcement learning engineerAdvanced control engineerRobotics engineerAI for control researcherIntelligent automation engineerRobotics simulation specialist

Admissions & scholarships

Requirements

University degree in engineering, science or architecture, or equivalent demonstrable professional experience.
Internet connection and a computer capable of running the programme software.
English sufficient to read technical standards.

Ready to start?

Send your application today. Our admissions team replies within 72 hours with your study plan and funding options.

Apply now Scholarships

FAQ

Frequently asked questions

The programme totals 1,900 hours over 12–14 months. You set your own pace within each intake, and your tutor helps you plan the workload around your job.

You work with Gymnasium, Stable-Baselines3, MuJoCo, NVIDIA Isaac Lab and ROS 2, using free, academic or evaluation versions. Your tutor indicates the free alternative wherever one exists.

A SEIUM Institute diploma, awarded on passing the continuous assessment and the Master's Thesis. It is SEIUM's own qualification, not an official state degree; its recognition depends on the programme completed and the regulations of each country.

Yes. Modules passed in a course or postgraduate diploma in the same area are recognised in the higher programme. Ask about the recognition policy when you enrol.

A university degree in engineering, science or architecture (or equivalent professional experience), a computer able to run the software, and English sufficient to read technical standards.

Start dates are set by SEIUM's faculty for each intake. Our admissions team will confirm the next available date when you apply.

Keep exploring

Related programmes

See all in Data, AI, Simulation & Analytics

Master'sS$14,200

Master’s in HPC & GPU for Simulation and AI

Online master's programme of 1,900 hours in HPC and GPU computing for simulation and AI: CUDA/OpenMP, scaling, performance and cost. 12 modules covering HPC architectures, parallel…

12–14 monthsOnlineData, AI, Simulation & Analytics

Master'sS$12,700

Master’s in Optimisation, Operations Research and Decision Engineering

Online master's programme of 1,900 hours in optimisation, operations research and decision-making: 12 modules covering modelling, linear, integer and non-linear programming, metaheuristics, constraint programming and scheduling,…

1900 hOnlineData, AI, Simulation & Analytics

Master'sS$8,450

Master’s in Visual Analytics and Technical BI

Online master's programme of 1,900 hours in visual analytics and technical business intelligence: dashboard design, data storytelling and operational decision-making. 12 modules covering BI fundamentals, visual…

12–14 monthsOnlineData, AI, Simulation & Analytics

Master’s in Reinforcement & Imitation Learning for ControlEnrol

Related programmes

More in Data, AI, Simulation & Analytics

Master’s in HPC & GPU Engineering for Simulation and AIView programme →Master’s in Industrial Data Engineering & Platform MLOpsView programme →Master’s in Synthetic Data and Sensor SimulationView programme →Master’s in Causality and Experimentation (A/B) for EngineeringView programme →Master’s in Visual Analytics and Technical BI EngineeringView programme →Master’s in Graph ML Engineering for Networks and SystemsView programme →

All Data, AI, Simulation & Analytics programmes → · How to apply · Fees & financing · Scholarships