|

Case Study Paper: Autonomous Driving

LIS 461: Data and Algorithms: Ethics and Policy

Instructor: Paul J. Kelly | Term: Summer 2024

Introduction

Autonomous vehicles are cars with advanced technology and multiple sensors that work with AI algorithms in the backend to process real-time data to navigate through roads safely and without human intervention. As AI innovation ramps up, individuals and corporations are starting to implement these algorithms into vehicle software to provide drivers with the ability to buy a car that could potentially drive itself.

These algorithms make decisions using data from multiple sensors throughout the vehicle, including LiDAR, radar, and more. Because they make real-time decisions on the road, they must weigh factors such as traffic laws and safety while respecting ethical principles like minimization of harm and respect for life.

Autonomous vehicles often face dilemmas where both outcomes can involve harmful consequences. This raises the Trolley Problem, a set of thought experiments about saving multiple lives by sacrificing one person. Car manufacturers and software companies should provide transparency to regulators and policymakers, including what data these systems are trained on and how bias may be incorporated into the technology.

Clarify Concepts

Autonomous vehicles rely on sensor data for algorithms to analyze and make real-time navigation decisions. These systems can interpret surroundings and generate driving actions, and some can improve through reinforcement learning. Companies worldwide continue to compete on autonomous-driving capability, including Tesla's Autopilot software.

Even with advanced software, driver supervision remains necessary in many current systems. The Society of Automotive Engineers (SAE) defines six levels of autonomy that range from no automation to full automation. Level 0 has no driving automation. Level 1 offers assistance such as cruise control-like support. Level 2, including examples like Tesla Autopilot, can perform some tasks such as steering, lane changes, and acceleration, but the driver must stay ready to take over.

Level 3 is conditional automation where the vehicle can handle driving under ideal conditions, but the driver must still be ready to intervene. Level 4 is high automation where intervention is usually unnecessary, though manual override is possible; the paper describes this level as primarily urban and low-speed use.

Waymo is discussed as a Level 4 example. In San Francisco, users can book rides in driverless Waymo vehicles, and as of June 25, 2024, its waitlist was removed for public access. The paper notes controversy around potential traffic impacts. Finally, Level 5 represents full driving automation in all conditions.

Facts Straight

Autonomous vehicles use algorithms that make key road decisions from sensor data, including LiDAR and radar, to perceive the surrounding environment. LiDAR (Light Detection and Ranging) uses laser pulses and optics to build a virtual 3D map of the car's surroundings, which can then be used by downstream algorithms to classify objects and predict their trajectories.

Latency, the delay between sensing and action, is a major safety concern. Vehicles need low-latency data for safe real-time decisions. If latency is high, the car may operate on outdated information. For this reason, the paper emphasizes local processing rather than cloud dependence for critical driving decisions.

The paper highlights FCW (Forward Collision Warning) as an essential safety feature. FCW uses cameras to monitor distance to the vehicle ahead and, when a collision risk is detected, can alert the driver and trigger automatic braking. Lane control is also identified as essential for keeping a vehicle centered and reducing unintended lane departure.

On regulation, the paper notes the lack of a comprehensive federal framework in the United States and that many policies are state-driven. Like all vehicles sold in the U.S., autonomous vehicles must pass FMVSS. The paper states that as of August 2024, 29 states and Washington, D.C. had passed legislation related to autonomous vehicles, with differences across states such as Texas, Arizona, and California.

Moral Theory

The paper compares Kantian ethics and utilitarianism in autonomous-driving decision-making. Kantian ethics emphasizes treating individuals as ends in themselves, not as means to an end. Applied to autonomous vehicles, this implies algorithms should avoid sacrificing one person solely to benefit others.

Utilitarianism focuses on outcomes and maximizing overall well-being. In this framework, algorithms may make trade-offs intended to reduce total harm, including choosing actions that could impose minor injuries on a few to prevent severe harm to many. The paper presents these frameworks as a core tension in policy design and moral evaluation for self-driving systems.

Policy Proposal

The paper proposes mandatory ethical review for autonomous vehicles entering the consumer market, including analysis of their algorithms and software by independent ethics boards. The goal is to ensure both Kantian and utilitarian considerations are addressed in deployment decisions.

It also calls for transparency and accountability measures, including documentation of how self-driving algorithms make decisions in ethically complex situations and clear responsibility frameworks for accidents or malfunctions. The proposal argues developers and manufacturers should bear at least equal accountability to operators when system behavior contributes to harm.

The paper addresses concerns that policy could stifle innovation or expose proprietary systems. It argues these requirements are necessary for long-term public trust and can be designed to protect sensitive information, including anonymization methods and transparency about decision factors rather than trade secrets.

Overall, the proposal aims to guard the moral integrity of autonomous vehicles while increasing safety and public confidence as deployment expands.

Bibliography

  1. Roff, Heather M. "The folly of trolleys: Ethical challenges and autonomous vehicles | Brookings." Brookings Institution, December 17, 2018. Source
  2. "SAE Levels of Driving Automation Refined for Clarity and International Audience." SAE International, May 3, 2021. Source
  3. Elias, Jennifer. "Waymo opens robotaxi service to all San Francisco users." CNBC, June 25, 2024. Source
  4. Watts, Michael, et al. "LIDAR ON A CHIP PUTS SELF-DRIVING CARS IN THE FAST LANE." IEEE Spectrum, 2023. Source
  5. Cicchino, Jessica. "Forward collision warning (FCW) alone, low-speed autonomous emergency braking (AEB), and FCW combined with AEB that operates at highway speeds reduced rear-end striking crash involvement rates by 27 percent, 43 percent, and 50 percent ..." ITS Deployment Evaluation, February 2, 2017. Source