NAVIGATION VLA RESEARCH FRAMEWORK

SimpleNav: Make Navigation VLA Simple.

A unified path from heterogeneous navigation data to composable models, reproducible training, and comparable evaluation.

OpenFly · TravelUAV · AerialVLN · R2R-CE · RxR-CE · EVT-Bench
Featured rolloutsA compact gallery of representative rollouts.

Open end to end: data, models, training, evaluation, and reproducible artifacts.

Research use only. Not a safety-certified flight-control system; validate with supervision, manual override, and independent safety monitors.

ONE RESEARCH LOOP

From raw trajectories to reliable decisions.

Select a layer to see what stays stable and where new datasets, models, or benchmarks plug in.

SimpleNAV data conversion, model training, and closed-loop evaluation framework
THE DATA LAYER

One format for heterogeneous navigation data.

SimpleNav converts heterogeneous navigation datasets into a shared LeRobot v3 format for training. With conversion, augmentation, and validation built in.

SimpleNav · convert · augment · validate

LeRobot v3

Unified training format

observation · state · action · episode
trajectories · metadata · multi-view video

THE MODEL LAYER

Simple on purpose. Modular underneath.

SimpleNAV includes its own long-horizon navigation architecture—not only a set of tools around a backbone.

SimpleNAV model architecture
RELEASE 01 / EVIDENCE

Results you can trace.

Switch benchmarks and splits to inspect headline results and complete method comparisons without a wide scrolling table.

ROLLOUT GALLERY

See representative rollouts.

A compact gallery of representative rollouts.

WHAT COMES NEXT

A framework that grows with the task.

The destination is not a fixed set of scripts. It is an extensible VLN foundation for different data, architectures, platforms, and deployment settings.

DOCUMENTATION

The source stays close to the work.

Stable topic documents separate public interfaces from run-specific notes, so the landing page stays readable as the project expands.