Autonomous Driving: AI Safety and Security

A structured, hands-on learning resource covering perception systems, functional safety, cybersecurity, and AI safety for autonomous vehicles — aligned with ISO 26262, ISO 21448, ISO/SAE 21434, and ISO/PAS 8800.

Author: Milin Patel · Hochschule Kempten — University of Applied Sciences

---

Table of Contents

• Overview
• Repository Structure
• Modules
  • 01 — Perception Systems
  • 02 — Failure Analysis
  • 03 — Functional Safety (ISO 26262)
  • 04 — SOTIF (ISO 21448)
  • 05 — Cybersecurity (ISO/SAE 21434)
  • 06 — AI Safety (ISO/PAS 8800)
  • 07 — Integration and Deployment
• Templates
• Learning Paths
• Getting Started
  • Prerequisites
  • Option A: Google Colab (Recommended)
  • Option B: Local Installation
• Standards and Regulations Covered
• Contributing
• Citation
• License

---

Overview

Autonomous vehicles depend on the safe and secure interaction of perception, decision-making, and control systems. This repository provides 33 Jupyter notebooks and 3 professional templates organized into 7 progressive modules that cover:

• Perception — How autonomous vehicles sense their environment (cameras, LiDAR, radar, sensor fusion)
• Failure analysis — Real-world incidents (Uber, Tesla, Cruise), adversarial attacks, and edge cases
• Functional safety — Systematic hazard analysis using ISO 26262 (HARA, FMEA, ASIL, V&V)
• SOTIF — Performance limitations and triggering conditions under ISO 21448
• Cybersecurity — Threat modeling and attack surface analysis per ISO/SAE 21434
• AI safety — Uncertainty quantification, calibration, and trustworthiness under ISO/PAS 8800
• Integration — V2X communication, explainability (XAI), regulations (UNECE R155/R156, EU AI Act)

Each notebook is self-contained, executable in Google Colab with a single click, and includes explanations, working code, visualizations, and exercises.

---

Repository Structure

Autonomous-Driving_AI-Safety-and-Security/
│
├── README.md
├── LICENSE
├── CONTRIBUTING.md
├── requirements.txt
├── .gitignore
│
├── 01_Perception_Systems/
│   └── notebooks/
│       ├── 01_sae_automation_levels.ipynb
│       ├── 02_sensor_technologies.ipynb
│       ├── 03_object_detection.ipynb
│       ├── 04_sensor_fusion.ipynb
│       ├── 05_pedestrian_detection.ipynb
│       ├── 06_lidar_sensor_fundamentals.ipynb
│       └── 07_dataset_overview.ipynb
│
├── 02_Failure_Analysis/
│   └── notebooks/
│       ├── 01_av_failure_case_studies.ipynb
│       ├── 02_ood_detection.ipynb
│       ├── 03_corner_cases_edge_cases.ipynb
│       └── 04_adversarial_attacks.ipynb
│
├── 03_Functional_Safety/
│   └── notebooks/
│       ├── 01_iso_26262_fundamentals.ipynb
│       ├── 02_hara_methodology.ipynb
│       ├── 03_fmea_analysis.ipynb
│       └── 04_verification_validation.ipynb
│
├── 04_SOTIF/
│   └── notebooks/
│       ├── 01_sotif_fundamentals.ipynb
│       ├── 02_scenario_analysis.ipynb
│       ├── 03_ood_detection_sotif.ipynb
│       └── 04_simulation_sotif_validation.ipynb
│
├── 05_Cybersecurity/
│   └── notebooks/
│       ├── 01_automotive_cybersecurity.ipynb
│       ├── 02_tara_methodology.ipynb
│       └── 03_attack_surface_analysis.ipynb
│
├── 06_AI_Safety/
│   └── notebooks/
│       ├── 01_ai_safety_standards.ipynb
│       ├── 02_uncertainty_types.ipynb
│       ├── 03_mc_dropout_ensembles.ipynb
│       ├── 04_calibration_reliability.ipynb
│       └── 05_safety_validation_testing.ipynb
│
├── 07_Integration_Deployment/
│   └── notebooks/
│       ├── 01_v2x_communication.ipynb
│       ├── 02_explainability_xai.ipynb
│       ├── 03_standards_integration.ipynb
│       ├── 04_industry_deployment.ipynb
│       ├── 05_odd_runtime_monitoring.ipynb
│       ├── 06_standards_gaps.ipynb
│       └── 07_regulations_type_approval.ipynb
│
└── templates/
    ├── HARA_Template.md
    ├── TARA_Template.md
    └── SOTIF_Analysis_Template.md

---

Modules

01 — Perception Systems

How autonomous vehicles perceive their surroundings through sensors and algorithms.

#	Notebook	Topic	Open
1	01_sae_automation_levels	SAE J3016 levels (L0–L5), ODD definition, system architecture
2	02_sensor_technologies	Camera, LiDAR, and radar — capabilities, limitations, trade-offs
3	03_object_detection	YOLO, Faster R-CNN, DETR — architectures and evaluation
4	04_sensor_fusion	Early, late, and deep fusion strategies for multi-sensor systems
5	05_pedestrian_detection	Specialized detection for vulnerable road users
6	06_lidar_sensor_fundamentals	3D point cloud processing and LiDAR data pipelines
7	07_dataset_overview	KITTI, nuScenes, and Waymo Open Dataset — structure and usage

---

02 — Failure Analysis

Learning from real-world failures and understanding robustness challenges.

#	Notebook	Topic	Open
1	01_av_failure_case_studies	Uber ATG (2018), Tesla Autopilot, Cruise incidents — root causes and lessons
2	02_ood_detection	Out-of-distribution detection — techniques and safety implications
3	03_corner_cases_edge_cases	Identifying and testing corner cases and edge cases
4	04_adversarial_attacks	Adversarial perturbations, physical attacks, and defenses

---

03 — Functional Safety (ISO 26262)

Systematic methods for identifying, classifying, and mitigating safety risks.

#	Notebook	Topic	Open
1	01_iso_26262_fundamentals	V-Model, safety lifecycle, ASIL determination
2	02_hara_methodology	Hazard Analysis and Risk Assessment — worked examples
3	03_fmea_analysis	Failure Mode and Effects Analysis for AV components
4	04_verification_validation	V&V strategies, test coverage, and compliance evidence

---

04 — SOTIF (ISO 21448)

Addressing performance limitations and insufficiencies of the intended functionality.

#	Notebook	Topic	Open
1	01_sotif_fundamentals	ISO 21448 principles, four scenario categories (S1–S4)
2	02_scenario_analysis	Triggering condition identification and scenario generation
3	03_ood_detection_sotif	Out-of-distribution detection applied to SOTIF validation
4	04_simulation_sotif_validation	Simulation-based testing and scenario coverage analysis

---

05 — Cybersecurity (ISO/SAE 21434)

Protecting autonomous vehicles against cyber threats across the development lifecycle.

#	Notebook	Topic	Open
1	01_automotive_cybersecurity	ISO/SAE 21434 framework, cybersecurity management systems
2	02_tara_methodology	Threat Analysis and Risk Assessment — step-by-step methodology
3	03_attack_surface_analysis	Attack surface identification, threat modeling, and mitigations

---

06 — AI Safety (ISO/PAS 8800)

Ensuring trustworthiness, reliability, and safety of AI/ML components.

#	Notebook	Topic	Open
1	01_ai_safety_standards	ISO/PAS 8800 framework, trustworthiness characteristics
2	02_uncertainty_types	Aleatoric vs. epistemic uncertainty in deep learning
3	03_mc_dropout_ensembles	MC Dropout, deep ensembles — uncertainty quantification
4	04_calibration_reliability	Confidence calibration, ECE, reliability diagrams
5	05_safety_validation_testing	Safety validation strategies for AI-based systems

---

07 — Integration and Deployment

Bringing safety, security, and AI together — from standards compliance to real-world deployment.

#	Notebook	Topic	Open
1	01_v2x_communication	Vehicle-to-Everything (V2X) communication and safety
2	02_explainability_xai	Explainable AI (XAI) for safety-critical decision-making
3	03_standards_integration	Integrating ISO 26262, ISO 21448, and ISO/SAE 21434
4	04_industry_deployment	Industry deployment challenges and operational readiness
5	05_odd_runtime_monitoring	ODD monitoring, runtime safety checks, and degradation
6	06_standards_gaps	Open problems and gaps in current safety/security standards
7	07_regulations_type_approval	UNECE R155/R156, EU AI Act, and type approval pathways

---

Templates

Ready-to-use analysis templates for professional safety and security work. Each template follows the structure defined in its reference standard, includes worked examples, and provides guidance for completing each section.

Template	Standard	Purpose
HARA Template	ISO 26262	Hazard Analysis and Risk Assessment — hazard identification, ASIL determination, safety goals
TARA Template	ISO/SAE 21434	Threat Analysis and Risk Assessment — asset identification, threat scenarios, CAL determination
SOTIF Analysis Template	ISO 21448	SOTIF evaluation — ODD definition, triggering conditions, scenario categorization (S1–S4)

---

Learning Paths

Choose a path based on your goals:

Complete path (all 33 notebooks):

01 Perception → 02 Failure Analysis → 03 Functional Safety → 04 SOTIF → 05 Cybersecurity → 06 AI Safety → 07 Integration

Perception and AI focus (16 notebooks):

01 Perception → 02 Failure Analysis → 06 AI Safety

Safety standards focus (19 notebooks):

03 Functional Safety → 04 SOTIF → 05 Cybersecurity → 07 Integration

Quick standards overview (4 notebooks):

03/01 ISO 26262 Fundamentals → 04/01 SOTIF Fundamentals → 05/01 Automotive Cybersecurity → 06/01 AI Safety Standards

---

Getting Started

Prerequisites

• Python 3.8 or later
• Basic familiarity with Python, NumPy, and Matplotlib
• Background in engineering, computer science, or automotive systems (helpful but not required)

Option A: Google Colab (Recommended)

Click any Open in Colab badge in the module tables above. No local setup is needed — each notebook installs its own dependencies automatically.

Option B: Local Installation

# 1. Clone the repository
git clone https://github.com/milinpatel07/Autonomous-Driving_AI-Safety-and-Security.git
cd Autonomous-Driving_AI-Safety-and-Security

# 2. Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Linux / macOS
venv\Scripts\activate           # Windows

# 3. Install dependencies
pip install -r requirements.txt

# 4. Launch Jupyter
jupyter lab

GPU support: For local GPU acceleration, install PyTorch with CUDA following the official instructions.

---

Standards and Regulations Covered

Safety and Security Standards

Standard	Scope	Modules
ISO 26262:2018	Functional safety of road vehicles	03, 07
ISO 21448:2022	Safety of the intended functionality (SOTIF)	04, 07
ISO/SAE 21434:2021	Cybersecurity engineering for road vehicles	05, 07
ISO/PAS 8800	Safety and artificial intelligence	06, 07
SAE J3016	Taxonomy of driving automation	01
ISO/IEC 24028	Trustworthiness in AI — overview	06

Regulatory Frameworks

Regulation	Scope	Module
UNECE R155	Cybersecurity management system requirements	07
UNECE R156	Software update management system requirements	07
EU AI Act	Risk-based regulation of artificial intelligence systems	07

---

Contributing

Contributions are welcome. Please read CONTRIBUTING.md for guidelines on how to contribute to this project, including how to report issues, suggest improvements, and submit pull requests.

---

Citation

If you use this material in academic or professional work, please cite:

@misc{patel2025av_safety,
  author       = {Patel, Milin},
  title        = {Autonomous Driving: AI Safety and Security},
  year         = {2025},
  publisher    = {GitHub},
  url          = {https://github.com/milinpatel07/Autonomous-Driving_AI-Safety-and-Security}
}

---

License

This project is licensed under the MIT License.

Copyright (c) 2025 Milin Patel