Postgraduate Diploma in RL/IL for Driving: Safety Layers and Reward Shaping
Online 800-hour Postgraduate Diploma in RL/IL for Driving: Safety Layers and Reward Shaping. 8 modules covering ISO/PAS 8800:2024, ISO 21448:2022 (SOTIF) and CARLA. Assigned tutor and open enrolment.
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.
What you will be able to do
Frame driving decision-making as a learning problem.
Apply imitation learning techniques.
Train policies with deep reinforcement learning.
Design robust reward functions.
Implement safety layers on learned policies.
Improve generalisation and sim-to-real transfer.
Validate learned policies in line with AI and safety standards.
Who it's for
- Reinforcement learning engineer who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- Machine learning engineer who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- Autonomous driving researcher who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- Planning engineer who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- AI safety engineer who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- Simulation engineer who wants to specialise in RL/IL for Driving: Safety Layers and Reward Shaping.
- Graduates in engineering or sciences seeking to enter the automotive sector.
What you'll work with
- ISO/PAS 8800:2024
- ISO 21448:2022 (SOTIF)
- ISO 26262:2018
- Reglamento (UE) 2024/1689 (Ley de IA)
- IEEE 2846-2022
- UL 4600
- ISO/IEC 22989:2022
8 modules · 800 h
1.1Limitations of rule-based programming1.2Markov decision processes1.3Observations, actions and rewards1.4Imitation versus reinforcement1.5End-to-end and modular driving1.6Simulators for learning1.7Documented successes and failures1.8Safety and generalisation challenges1.9Partial observability and POMDP1.10Practical case: analysis of a published end-to-end system
2.1Behavioural cloning2.2Distribution shift and DAgger2.3Inverse reinforcement learning2.4GAIL2.5Quality of demonstrations2.6Command-conditioned policies2.7Imitation with fleet data2.8Closed-loop evaluation2.9Causal confusion in imitation2.10Lab: behavioural cloning in CARLA
3.1Value-based methods and DQN3.2Policy gradient and PPO3.3Actor-critic: SAC and TD33.4Model-based RL3.5Offline RL with CQL and IQL3.6Sample efficiency3.7Parallelisation of environments3.8Hyperparameters and stability3.9Continuous and discrete action spaces3.10Lab: PPO in highway-env
4.1Progress, safety and comfort4.2Potential-based reward shaping4.3Sparse and dense rewards4.4Reward hacking4.5Rewards learned from preferences4.6Penalties for infractions4.7Multi-objective rewards4.8Sensitivity analysis4.9Reward normalisation and scaling4.10Practice: comparison of reward functions
5.1Constrained RL and CMDP5.2Safety shields5.3Control barrier functions5.4Safe action projection5.5RSS as a safety layer5.6Reachability analysis5.7Simplex architectures5.8Policy monitoring5.9Safe exploration during training5.10Lab: safety filter with CBF
6.1Sim-to-real gap6.2Domain randomisation6.3Curriculum learning6.4Adversarial scenario generation6.5Policies robust to perturbations6.6Out-of-distribution evaluation6.7World models6.8Fine-tuning with real data6.9Overfitting to the simulator6.10Practical case: transferring a policy between cities
7.1Safety metrics in simulation7.2Failure search in critical scenarios7.3Neural network verification7.4Policy explainability7.5ISO/PAS 8800 requirements7.6AI Act and high-risk systems7.7Traceability of data and models7.8Model approval7.9Policy version registry7.10Workshop: validation plan for a learned policy
8.1Task and environment8.2Demonstration or simulation data8.3Reward function8.4Policy training8.5Safety layer8.6Scenario evaluation8.7Generalisation analysis8.8Risk discussion8.9Report and presentation8.10Final review with the tutor
How you'll learn
100% online training on the SEIUM campus, with downloadable material, recorded classes and assigned technical tutor. There are no fixed timetables: students progress at their own pace within the intake, which makes it possible to combine the programme with work.
Each module combines theory, guided reading of the original standard and applied exercises. The working software is CARLA, Stable-Baselines3, highway-env, PyTorch, Ray RLlib, in free, academic or evaluation versions. The tests are studied using real policy-training procedures in highway-env, behavioural cloning in CARLA, and integration of a safety filter over a policy. Assessment is continuous per module plus the final integrating project.
Build it end to end
Your final project follows the real engineering workflow, step by step, reviewed by your tutor.
Frame driving decision-making as a learning problem.
Apply imitation learning techniques.
Train policies with deep reinforcement learning.
Design robust reward functions.
Implement safety layers on learned policies.
Where this takes you
Requirements
- Prior technical training or professional experience in the sector. No university degree is required.
- Internet connection and a computer capable of running the programme's software.
- Sufficient level of Spanish to follow technical regulatory documentation.
Ready to start?
Send your application today. Our admissions team replies within 72 hours with your study plan and funding options.
Frequently asked questions
The programme consists of 800 hours in total, spread over 6 to 8 months. That is between 10 and 15 hours per week on average, compatible with a full-time job.
The programme works with CARLA, Stable-Baselines3, highway-env, PyTorch and Ray RLlib. Free, academic or evaluation versions are used, and the tutor indicates the free alternative where one exists.
An own diploma from SEIUM upon passing the continuous assessment and the final integrating project. It is in-house training, not an official qualification from the Ministry.
Yes. Modules passed in a course or postgraduate diploma in the same area are recognised in the advanced programme. Check the recognition policy when you enrol.
Related programmes
Postgraduate Diploma in Accelerator-Aware Architectures and Performance Measurement
Online Postgraduate Diploma of 800 h in Accelerator-Aware Architectures and Performance Measurement. 8 modules with ISO 26262:2018, ISO/PAS 8800:2024 and NVIDIA Nsight Compute. Assigned tutor and…
Postgraduate Diploma in Multimodal Fusion and Cross-Calibration
Online 800-hour Postgraduate Diploma in Multimodal Fusion and Cross-Calibration. 8 modules covering ISO 26262:2018, ISO 21448:2022 (SOTIF) and ROS 2. Assigned tutor and open enrolment.
Postgraduate Diploma in Secure Partitioning, ASIL and Inter-VM Communication
Online 800-hour Postgraduate Diploma in Secure Partitioning, ASIL and Inter-VM Communication. 8 modules covering ISO 26262:2018, ISO/SAE 21434:2021 and Ansys medini analyze. Assigned tutor and open…
More in Autonomous, ADAS & Software-Defined Vehicle
All Autonomous, ADAS & Software-Defined Vehicle programmes → · How to apply · Fees & financing · Scholarships