paper/arxiv/2609.02341

Robotics Dataset

This research paper studies multi-dataset training for Vision-Language-Action models in autonomous driving, introducing BEV-Forcing, an auxiliary objective that transfers bird's-eye-view object-layout information to improve zero-shot transfer across unseen datasets and camera rigs.

Catalog metadata

Repository owner

Caio Azevedo, Stefano Sabatini, Sascha Hornauer et al.

Source

Research paper

Task

Towards Zero-Shot Transfer Across Embodiments For Driving VLAs

Cameras

0

Browse the catalog to compare robots, tasks, sensors, and licenses. Sign in to open protected download and model links.