Automated Synthesis of Certifiable Control Software for Autonomous Vehicles

Introduction

Autonomous driving is a dominant technological theme of the 21st century, with vehicles from various car manufacturers equipped with different levels of autonomy. However, the implementation of safe and reliable control software remains a critical challenge for car manufacturers before fully autonomous vehicles become a reality.

Challenges in Current Autonomous Vehicle Software

The current lack of reliability in autonomous vehicle software is mainly attributed to the absence of formal correctness. Design requirements are not expressed in a formal language, leading to ambiguity and false implementations.

Human Factors and Testing Limitations

Furthermore, human factors from engineers to developers increase the likelihood of errors. Additionally, testing scenarios are limited, leaving many edge cases untested. Immature control software leading to accidents undermines public trust in autonomous vehicles and impedes further development.

Proposed Solution

We propose leveraging our ongoing ERC project AutoCPS research to create robust, reliable, and formally correct automotive control software. CertiCar proposes a software development framework that automates control software generation and guarantees its robustness and reliability.

Importance of Advanced Collision Avoidance Systems

A reliable advanced collision avoidance system (ACAS) is a fundamental requirement for future autonomous vehicles, significantly reducing the number of accidents, injuries, and fatalities.

Project Goals

CertiCar's PoC project aims to provide a correct-by-design ACAS with guaranteed correctness, reducing testing time by several orders of magnitude for car companies. This represents an initial step toward a complete software stack for a certifiable autonomous car.