GNM: A General Navigation Model to Drive Any Robot (Summary Video)

We present a General Navigation Model that can control any robot without fine-tuning or additional data collection. We find that the GNM outperforms policies trained on individual datasets and can also control new robots, such as a quadrotor, despite seeing no training data on aerial robots.
"GNM: A General Navigation Model to Drive Any Robot"
Dhruv Shah*, Ajay Sridhar*, Arjun Bhorkar, Noriaki Hirose, Sergey Levine
Berkeley AI Research (UC Berkeley), Toyota Motor North America
Project Page: sites.google.c...
arXiv pre-print: arxiv.org/abs/...
Timeline:
--------------
00:00 Introduction
00:17 Problem Setup
00:49 Training Data
01:10 Method Overview
03:38 Real-world Deployment
04:50 Robustness to Degradation
06:00 The End