RT-1: Robotics Transformer for Real-World Control at Scale

Presents Robotics Transformer, which exhibits promising scalable, pre-trained model properties for real-world control tasks.

This paper presents a model class called Robotics Transformer that can be trained on large, diverse, task-agnostic datasets, which can solve specific downstream tasks either zero-shot or with small task-specific datasets to a high level of performance. The authors argue that the success of such general robotic models lies with open-ended task-agnostic training, combined with high-capacity architectures that can absorb all of the diverse, robotic data.