Abstract:Developing safe autonomous driving systems is a major scientific and technical challenge. Existing AI-based end-to-end solutions do not offer the necessary safety guarantees, while traditional systems engineering approaches are defeated by the complexity of the problem. Currently, there is an increasing interest in hybrid design solutions, integrating machine learning components, when necessary, while using model-based components for goal management and planning. We study a method for building safe by design autonomous driving systems, based on the assumption that the capability to drive boils down to the coordinated execution of a given set of driving operations. The assumption is substantiated by a compositionality result considering that autopilots are dynamic systems receiving a small number of types of vistas as input, each vista defining a free space in its neighborhood. It is shown that safe driving for each type of vista in the corresponding free space, implies safe driving for any possible scenario under some easy-to-check conditions concerning the transition between vistas. The designed autopilot comprises distinct control policies one per type of vista, articulated in two consecutive phases. The first phase consists of carefully managing a potentially risky situation by virtually reducing speed, while the second phase consists of exiting the situation by accelerating. The autopilots designed use for their predictions simple functions characterizing the acceleration and deceleration capabilities of the vehicles. They cover the main driving operations, including entering a main road, overtaking, crossing intersections protected by traffic lights or signals, and driving on freeways. The results presented reinforce the case for hybrid solutions that incorporate mathematically elegant and robust decision methods that are safe by design.

What problem does this paper attempt to address?

This paper discusses how to design secure autonomous driving systems. Currently, AI-based end-to-end solutions lack the necessary security guarantees, while traditional systems engineering approaches struggle with complexity issues. The study presents a hybrid design approach that combines machine learning components (where necessary) and model-based components (for goal management and planning) to address the safety challenges of autonomous driving. The method is based on the assumption that driving capability can be attributed to the coordinated execution of a set of driving maneuvers. The study validates this assumption through a compositional result about viewpoints, where each viewpoint defines the free space in its vicinity. If safe driving is ensured within each type of viewpoint, then safe driving can be achieved in any possible scenario that satisfies some easily checkable conditions. The designed autonomous vehicle includes independent control strategies for each type of viewpoint, divided into two consecutive stages: the first stage cautiously handles potential hazardous situations through virtual deceleration, and the second stage exits such situations through acceleration. The autonomous vehicle utilizes simple functions to predict the vehicle's acceleration and deceleration, covering major driving maneuvers such as merging onto a main road, overtaking, navigating signalized intersections, and driving on highways. The results support a hybrid solution that adopts mathematically elegant and robust decision-making methods, which consider safety during the design phase. The paper points out the growing interest in autonomous driving systems, which integrate the advantages of both approaches - using machine learning for perception when needed, and model-based methods for goal management and planning. The study also emphasizes the issues of correctness (if control strategies are safe, the entire autonomous driving behavior is safe) and completeness (the control strategy set is sufficient for safe driving in all situations). Finally, the paper presents a viewpoint-based control strategy that achieves overall autonomous driving safety by ensuring safe driving within different types of viewpoints. This approach reduces design complexity and offers a new perspective for building secure autonomous driving systems.

Safe by Design Autonomous Driving Systems

A Shared Control Approach for Autonomous Vehicles via Driver Behaviors Learning

Perspectives on the system-level design of a safe autonomous driving stack

Correct by Design Coordination of Autonomous Driving Systems

Ensuring Safe Autonomy: Navigating the Future of Autonomous Vehicles

Integrated Modular Safety System Design for Intelligent Autonomous Vehicles

A formal framework for the safe design of the Autonomous Driving supervision

A Safe Control Architecture Based on a Model Predictive Control Supervisor for Autonomous Driving

SafeDrive: Knowledge- and Data-Driven Risk-Sensitive Decision-Making for Autonomous Vehicles with Large Language Models

Road to safe autonomy with data and formal reasoning

End-to-End Urban Autonomous Driving With Safety Constraints

Autonomous Vehicle Safety: An Interdisciplinary Challenge

How to Guarantee Driving Safety for Autonomous Vehicles in a Real-World Environment: A Perspective on Self-Evolution Mechanisms

Safe Real-World Autonomous Driving by Learning to Predict and Plan with a Mixture of Experts

Safety Co-Pilot: A System for Autonomous Vehicle to Make Decision Safer and Smarter

Safe and Generalized end-to-end Autonomous Driving System with Reinforcement Learning and Demonstrations

Hierarchical Control Strategy Towards Safe Driving of Autonomous Vehicles

Towards Safe, Explainable, and Regulated Autonomous Driving

Reachability-Based Decision-Making for Autonomous Driving: Theory and Experiments

An Algorithm for Supervised Driving of Cooperative Semi-Autonomous Vehicles (Extended)

A Holistic Safe Planner for Automated Driving Considering Interaction with Human Drivers