Autonomous Systems & Self-Driving AI

🚗

Inference Specialization

Autonomous systems engineers build the perception, prediction, and planning stacks for self-driving vehicles and drones. Massive investment from Waymo, Tesla, Mobileye, Aurora. One of the most technically demanding fields in applied AI.

Days

Projects

$200K

Median

$500B+

Market by 2030

$145K–$280K

Salary Range

Week 1: Sensor Processing and Perception Stack · Days 1–10  

Master the sensor modalities — LiDAR, camera, radar — and the algorithms that fuse them for robust 3D scene understanding.

D1–3

Autonomous vehicle architecture, sensor stack, and data formats

LiDAR, camera, radar, GNSS/IMU system architecture. Sensor calibration, synchronisation, and nuScenes format

↗ Self-Driving Cars Specialization — UoT/Coursera (free audit)

D4–6

3D object detection: PointPillars, VoxelNet, and BEV fusion

LiDAR-based 3D detection, bird's eye view camera-LiDAR fusion, evaluation on nuScenes and KITTI benchmarks

↗ Self-Driving Cars C2: State Estimation — Coursera (free audit)

D7–9

Lane detection, drivable area segmentation, and HD mapping

Lane segmentation networks, road surface classification, building HD maps from multi-sensor fusion data

↗ Self-Driving Cars C3: Visual Perception — Coursera (free audit)

🏗 Project — Day D10: 3D Object Detection System on LiDAR Point Clouds 

· PointPillars model trained on KITTI dataset

· mAP evaluation at IoU 0.5 and IoU 0.7 thresholds

· Real-time inference running at 10 Hz

· ROS 2 visualisation node with 3D bounding boxes

Week 2: Prediction, Planning, and Closed-Loop Simulation · Days 11–28  

Predict agent behaviour, plan safe trajectories, and test the full stack end-to-end in the CARLA simulator.

D11–14

Motion prediction: LSTM, Transformer, and social-force models

Pedestrian trajectory prediction, vehicle intent recognition, multi-modal probabilistic trajectory output

↗ Self-Driving Cars C4: Motion Planning — Coursera (free audit)

D15–18

Path planning: A-star, RRT-star, lattice planners, and MPC

Global planning with A-star, local reactive planning with RRT-star, Model Predictive Control for tracking

↗ Principles of Autonomy — MIT OCW (free lectures)

D19–22

CARLA simulator: closed-loop testing and adversarial scenarios

CARLA setup, scenario runner, adversarial agent injection, coverage metric tracking for test completeness

↗ CARLA Simulator official documentation (free)

D23–27

End-to-end learning vs modular pipelines: UniAD and VAD

Implicit affordance networks, UniAD unified AD model, comparing E2E and modular architectures

↗ UniAD paper — arXiv (free)

🏗 Project — Day D28: Closed-Loop Autonomous Driving Simulation Pipeline 

· CARLA simulation with 3 distinct driving scenarios

· Perception, prediction, and planning pipeline integrated

· Collision rate and ride comfort score metrics

· Failure mode analysis covering 5 edge cases

Week 3: Functional Safety, Validation, and Drone AI · Days 29–40  

ISO 26262 functional safety requirements, systematic test coverage validation, and aerial autonomous systems.

D29–32

Functional safety: ISO 26262, SOTIF, and safety case methodology

Automotive functional safety standard tiers. Safety of Intended Functionality, safety case and argument structure

↗ Self-Driving Car Engineer Nanodegree — Udacity (free preview)

D33–36

Validation: corner case mining and scenario-based test coverage

Data-driven coverage metrics, adversarial scenario generation, simulation-to-real transfer validation studies

↗ CARLA scenario runner and documentation (free)

D37–39

Drone AI: MAVLink, PX4 autopilot, and aerial perception

Drone flight control stack, PX4 autopilot integration, aerial 3D object detection, obstacle avoidance for UAVs

↗ PX4 Autopilot official documentation (free)

🏗 Project — Day D40: Autonomous System — Full-Stack Safety-Validated Capstone 

· Complete AV stack in CARLA: perception, prediction, and planning

· ISO 26262 hazard analysis and risk assessment document

· 50-scenario test suite with individual pass/fail records

· Corner case discovery using adversarial scenario injection

· Architecture diagram and safety case argument write-up