Arguing Machines: Perception-Control System Redundancy and Edge Case Discovery in Real-World Autonomous Driving

  • 2017-10-12 11:27:08
  • Lex Fridman, Benedikt Jenik, Bryan Reimer
  • 18

Abstract

Safe autonomous driving may be one of the most difficult engineeringchallenges that any artificial intelligence system has been asked to do sincethe birth of AI over sixty years ago. The difficulty is not within the taskitself, but rather in the extremely small margin of allowable error given thehuman life at stake and the extremely large number of edge cases that have tobe accounted for. In other words, we task these systems to expect theunexpected with near 100% accuracy, which is a technical challenge for machinelearning methods that to date have generally been better at memorizing theexpected than predicting the unexpected. In fact, the process of efficientlyand automatically discovering the edge cases of driving may be the key tosolving this engineering challenge. In this work, we propose and evaluate amethod for discovering edge cases by monitoring the disagreement between twomonocular-vision-based automated steering systems. The first is a proprietaryTesla Autopilot system equipped in the first generation of Autopilot-capablevehicles. The second is a end-to-end neural network trained on a large-scalenaturalistic dataset of 420 hours or 45 million frames of autonomous driving inTesla vehicles.

 

Quick Read (beta)

loading the full paper ...