ViNT
Visual Navigation Transformer — foundation model for robot navigation
ViNT is to mobile robots what GPT is to text. It was trained on footage from dozens of different robots navigating all kinds of spaces, and it learned a general 'sense of how to move.' Put a new robot in a new place and it can navigate without specific programming.
One of the most accessible foundation models in robotics. Lightweight, fast, Jetson-compatible, and actively maintained. If you have a mobile robot, this is worth trying.
Foundation model for visual navigation using a Transformer architecture. Learns navigational affordances from diverse cross-embodiment data and generalizes to novel environments.
GNM dataset (cross-embodiment, multiple robot types and environments)
Navigate a new warehouse layout without remapping
Last-mile outdoor navigation in novel environments
Autonomous field navigation with visual goals
Requires some ML and Python experience
RELATED DISCUSSIONS
Community →Seeded and planned prompts — not live forum activity yet.
- Community signalIntermediate / AdvancedHow should EAR track open robotics models?
Meta-discussion on ontology, benchmarks, and graph links for open-weight VLAs and manipulation models.
- Community signalIntermediate / AdvancedOpenVLA in the field — what breaks first?
Open Models room prompt linked to openvla-7b — inference, fine-tuning, and sim-to-real gaps. Seeded prompt only.