InternNav v0.3.0 Release

Changelog of v0.3.0 (2026/01/05)

Highlights

• Support training of InternVLA-N1 and evaluation on RxR (#184)
• Support training and evaluation for the VL-LN benchmark (#193, #198)
• Add a new flash without collision controller (#189)

New Features

• Add training code for InternVLA-N1 (#184)
• Support evaluation on the RxR dataset (#184)
• Add training code for the VL-LN benchmark baseline (#198)
• Support evaluation on VL-LN benchmark (#193)
• Add a Flash-without-Collisoin controller (#189)

Improvements

• Decouple System 2 and Dual-System evaluation functions in the Habitat evaluator for better readability (#184)
• Update InternVLA-N1 agent in VLN-PE to align with the updated InternVLA-N1 policy interface (#184)
• Enhance the Habitat evaluation pipeline to handle NaN values in results (#217)
• Update the README to include community tutorials (#217)

Bug Fixes

• Fix the version of diffusers in the requirements (#184)
• Fix the result JSON saving path in VLN-PE (#217)
• Fix a bug in RxR evaluation result collection (#217)
• Removed legacy code in scripts/demo (#217)

Contributors

@kellyiss @DuangZhu @0309hws @kew6688

Full Changelog: release/v0.2.0...release/v0.3.0