GeoDrive: 3D Geometry-Informed Driving World Model with Precise Action Control
Anthony Chen, Wenzhao Zheng, Yida Wang, Xueyang Zhang, Kun Zhan, Peng Jia, Kurt Keutzer, Shanghang Zhang
2025-05-30
Summary
This paper talks about GeoDrive, a new system for self-driving cars that uses detailed 3D information to help the car understand its surroundings and make safer, more accurate driving decisions.
What's the problem?
The problem is that many self-driving car systems struggle to fully understand the space around them, which can lead to mistakes in navigation and controlling the car, especially in tricky or crowded environments.
What's the solution?
The researchers built GeoDrive to include strong 3D geometry knowledge in the car's world model. This means the car gets a much clearer and more accurate picture of where things are and how to move around them, which helps it control its actions more precisely.
Why it matters?
This is important because it makes self-driving cars safer and more reliable, which is crucial for earning people's trust and making autonomous vehicles a real option for everyday transportation.
Abstract
GeoDrive integrates robust 3D geometry into driving world models to improve spatial understanding and action controllability in autonomous navigation, enhancing safety and reliability.