InternNav v0.2.0 Release

Changelog of v0.2.0 (2025/12/4)

Highlights

• Support distributed evaluation for VLN-PE, reducing full benchmark runtime to ~1.6 hours using 16 GPUs (≈13× speedup over single-GPU eval) (#168)
• Enhance Habitat evaluation flow with DistributedEvaluator and HabitatEnv integrated into the InternNav framework (#168)
• Support install flags for dependency isolation: [habitat], [isaac], [model] (#135)

New Features

• Support distributed evaluation for VLN-PE (#168)
• Support a unified evaluation script eval.py, with new Habitat evaluation configs in scripts/eval/configs (#168)
• Support install flags for dependency isolation (#168)

Improvements

• Add HabitatEnv with episode pool management (#168)
• Update InternUtopiaEnv for distributed execution and episode pool management (#168)
• Enhance episode_loader in VLN-PE with new distributed mode compatibility (#168)
• Update data_collector to support progress checkpointing and incremental result aggregation in distributed evaluation. (#168)

Bug Fixes

• Fix logger disabled after Isaac Sim initialization during evaluator bootstrap (#168)
• Fix dataloader bug where revise_one_data() was incorrectly applied to all datasets (#168)
• Fix visualization images dimension mismatch during InternVLA-N1 evaluation (#168)
• Fix distributed evaluation crash in rdp policy (#168)
• Fix GitHub CI tests (#168)

Contributors

A total of 3 developers contributed to this release.
@kew6688, @Gariscat, @yuqiang-yang

Full changelog: release/v0.1.0...release/v0.2.0