jishnuperiya/nd0013-self-driving-car-engineer

This is the repository for the ND0013 - Self-Driving Car Engineer Nanodegree programme given at Udacity during the 2022 session.

Self-Driving Car Engineer Nanodegree

Udacity ND0013 | 2022 Cohort

This is the repository for the ND0013 - Self-Driving Car Engineer Nanodegree programme given at Udacity during the 2022 session.

Courses

• ✅ Course 1: Computer Vision
• ✅ Course 2: Sensor Fusion
• ✅ Course 3: Localization
• ✅ Course 4: Planning
• ✅ Course 5: Control

Projects

• ✅ Project 1.1: Object Detection in Urban Environments (2D)
• ✅ Project 2.1: 3D Object Detection with LiDAR Data
• ✅ Project 2.2: Multi-Target Tracking with Extended Kalman filter (MTT with EKF)
• ✅ Project 3.1: Scan Matching Localization
• ✅ Project 4.1: Motion Planning and Decision Making for Autonomous Vehicles
• ✅ Project 5.1: Control and Trajectory Tracking for Autonomous Vehicles

Contents

The following topics are covered in course projects:

Course 1: Computer Vision

• Training a SSD object detection model on the Waymo Open Dataset in TensorFlow
• Fine-tuning strategies and making architectural optimisations for DNNs
• Conducting exploratory data analysis (EDA) and training / evaluation error analysis
• And so much more ... (see 1-Computer-Vision for full list of course topics).

Course 2: Sensor Fusion

• Extract and transform LiDAR range images into 3D point clouds and bird's-eye view (BEV) maps;
• Build and experiment with state-of-the-art 3D object detection nets;
• Pre-process and perform multi-object tracking with multi-modal sensor data (LiDAR, RGB camera, radar);
• Implement the Extended Kalman filter (EKF) and the Unscented Kalman filter (UKF) for multi-object tracking;
• And so much more ... (see 2-Sensor-Fusion for full list of course topics).

Course 3: Localization

• Master robot localisation from one-dimensional motion models to three-dimensional point cloud maps;
• Master the fundamentals of Bayes' theorem and the Markov assumption applied to robot localisation;
• Implement Markov localisation to perform 1D object tracking in C++;
• Write and optimise two scan matching algorithms in C++: Iterative Closest Point (ICP) and Normal Distributions Transform (NDT);
• Apply the scan matching algorithms to simulated LiDAR point clouds processed with the Point Cloud Library (PCL);
• And so much more ... (see 3-Localization for full list of course topics).

Course 4: Planning

• Design and implement weighted cost functions and behaviour planning systems in C++;
• Perform structured trajectory generation in C++;
• Implement the A* and Hybrid A* search algorithms in C++;
• Use numerical approximation and discretisation to solve the Polynomial splines problem;
• Generate optimal, feasible, collision-free paths;
• And so much more ... (see 4-Planning for full list of course topics).

Course 5: Control

• Design and implement feedback controllers (PID and MPC) for trajectory tracking;
• Select design parameters to guarantee stability;
• Use MPC to design a feedback controller for non-linear dynamics;
• Test and evaluate the feedback controllers w.r.t. real-world perturbations using CARLA Simulator;
• And so much more ... (see 5-Control for full list of course topics).

Material

Syllabus:

• Program Syllabus | Udacity Nanodegree

Literature:

• See specific courses for related literature.

Datasets:

• Waymo Open Dataset: Perception;
• German Traffic Sign Recognition Benchmark GTSDB.

Lectures:

• Lecture materials (videos, slides) available offline. Course lecture notes available on request.

Other resources

Companion code:

• Object Detection in an Urban Environment | Starter code;
• Sensor Fusion and Tracking | Starter code;
• SFND Lidar Obstacle Detection | Starter code;
• Control and Trajectory Tracking | Starter code.