What humanlike errors do autonomous vehicles need to avoid to maximize safety?

Area of Science:

• Traffic Safety
• Human-Computer Interaction
• Autonomous Systems

Background:

• Driver error is the primary cause of 94% of traffic crashes.
• Autonomous vehicles (AVs) are expected to reduce crashes but may still err like human drivers.
• Understanding crash contributing factors is crucial for AV safety development.

Purpose of the Study:

• To analyze crash data and identify factors that autonomous vehicles (AVs) must address to maximize safety.
• To determine the potential crash reduction achievable by AVs, considering human-like error patterns.

Main Methods:

• Utilized the National Motor Vehicle Crash Causation Survey (NMVCCS) database.
• Categorized driver-related crash factors into sensing/perceiving, predicting, planning/deciding, execution/performance, and incapacitation.
• Modeled crash persistence assuming AVs overcome incapacitation and perception limitations.

Main Results:

• 33% of crashes involved only perception or incapacitation factors, potentially preventable by AVs.
• 67% of crashes could persist, with significant contributions from planning/deciding (41%), execution/performance (23%), and predicting (17%) factors.
• Planning/deciding errors frequently involved speeding and illegal maneuvers.

Conclusions:

• AVs must avoid replicating human errors in decision-making, prediction, and control to achieve safety goals.
• AV design must prioritize safety over occupant preferences, especially concerning traffic law adherence.
• Addressing planning/deciding factors is critical for AVs to realize their crash elimination potential.