The Neural Network That Drove Across America
A minivan steered by a neural network crossed 2,849 miles of American highway, proving self-driving cars were not science fiction.
People thought we were crazy. But we wanted to show that neural networks could handle the real world, not just toy problems.
— Dean Pomerleau
In 1995, Dean Pomerleau and Todd Jochem’s No Hands Across America trip demonstrated the potential of autonomous driving technology, showcasing a neural network’s ability to steer a vehicle across the United States.
| What happened: In 1995, Dean Pomerleau and Todd Jochem undertook the No Hands Across America project, which involved driving a modified Navlab 5 vehicle from Pittsburgh to San Diego. The vehicle was equipped with ALVINN (Autonomous Land Vehicle In a Neural Network), a pioneering neural network system that could interpret road images and steer the vehicle autonomously. This journey covered over 3,000 miles, with the neural network successfully steering the vehicle for 98.2% of the trip. Carnegie Mellon | ALVINN |
Why it matters: This groundbreaking demonstration laid the conceptual groundwork for modern self-driving systems, proving that neural networks could effectively handle complex driving tasks. It predates the efforts of companies like Tesla and Waymo by decades and has had a lasting impact on the development of autonomous vehicle technology.
Further reading:
Why This Mattered
The No Hands Across America trip demonstrated that a neural network trained on road images could steer a vehicle for 98.2% of a coast-to-coast journey. It was the longest autonomous drive ever attempted at the time and laid conceptual groundwork for modern self-driving systems decades before Tesla or Waymo existed.
