OpenBMB/SimpleNav

OpenBMB/SimpleNav

Description: SimpleNav: A simple, unified, reproducible, and extensible framework for navigation VLA research.

Language: Python

License: MIT

Stars: 11

Forks: 0

Open issues: 0

Created: 2026-08-27T02:52:20Z

Pushed: 2026-08-28T02:46:44Z

Default branch: main

Fork: no

Archived: no

README:

SimpleNav

Make Navigation VLA Simple.

A simple, unified, reproducible, and extensible framework for navigation VLA research.

中文 · Project Page · Data Pipeline · Documentation · Results · Data, Environments & Models

SimpleNav is a simple, unified, reproducible, and extensible framework for navigation VLA research, jointly developed and open-sourced by THUNLP at Tsinghua University, AI9Stars, OpenBMB, and HITDIP. It provides a unified research pipeline that connects heterogeneous navigation data, long-horizon VLA models, training, and benchmark evaluation through well-defined interfaces. SimpleNav supports both aerial and ground navigation, preserves dataset-specific coordinate systems and simulator semantics through dedicated adapters, and standardizes model, action, artifact, and evaluation interfaces to enable efficient reuse, comparison, and extension across datasets, tasks, and platforms.

Table of Contents

• [SimpleNav](#simplenav)
• [Vision](#vision)
• [Why SimpleNav](#why-simplenav)
• [Framework](#framework)
• [Data Protocol](#data-protocol)
• [Trajectory augmentation](#trajectory-augmentation)
• [Model](#model)
• [Results](#results)
• [Demos](#demos)
• [Risks and Limitations](#risks-and-limitations)
• [Quick Start](#quick-start)
• [1. Clone and install the model environment](#1-clone-and-install-the-model-environment)
• [2. Prepare data](#2-prepare-data)
• [3. Train](#3-train)
• [4. Evaluate](#4-evaluate)
• [Documentation](#documentation)
• [Roadmap](#roadmap)
• [Citation](#citation)
• [License](#license)
• [Acknowledgements](#acknowledgements)

Vision

Navigation research should not require a separate data-model-evaluation stack for every dataset. SimpleNav provides one research loop in which:

• source datasets enter through explicit conversion adapters;
• model components remain replaceable and composable;
• training runs are defined by portable configs;
• benchmark-specific behavior stays inside evaluation plugins;
• results remain traceable to code, data, config, checkpoint, and simulator versions.

Why SimpleNav

| Area | What is provided | | --- | --- | | Simple Data | Conversion, trajectory augmentation, AirSim image collection, LeRobot v3 writing, validation, statistics, BATS context, and visual-token cache tools. | | Simple Model | Qwen3.5-VL navigation, long-history selection, temporal-view encoding, visual-token caching, and diffusion action heads. | | Simple Training | Configuration-driven local, distributed, single-dataset, and mixed-dataset training. | | Simple Evaluation | Portable OpenFly, TravelUAV, AerialVLN, EVT-Bench, R2R-CE, and RxR-CE configs with shared rollout artifacts. |

Framework


| Path | Description | | --- | --- | | [data_pipeline/](data_pipeline/README.md) | Raw-data conversion, trajectory augmentation, simulator image collection, and enhanced-data construction. | | [starVLA/](starVLA/) | Dataloaders, models, training runtime, and shared modules. | | [examples/NavVLA/](examples/NavVLA/) | Portable training entry points and configs. | | [NavVLAeval/](NavVLAeval/README.md) | Closed-loop and offline benchmark evaluation. | | [tool/navvla/](tool/navvla/README.md) | Dataset validation, repair, statistics, context, cache, and open-loop tools. | | [deployment/](deployment/) | Deployment-side entry points. | | [docs/](docs/guides/README.md) | Documentation for installation, data, models, training, evaluation, and results. |

Data Protocol

The primary LeRobot dataloader keeps storage, model input, and prediction target separate:

| Field | Protocol | | --- | --- | | Stored observation.state | One pose [x, y, z, yaw] in the coordinate convention declared by the dataset adapter. | | Model state | When include_state: true, consecutive body-frame relative motions over the selected BATS history, ending at the current frame. It is not the stored absolute pose or the future action target. | | Primary action target | A future chunk [H, 4] of [dx_forward, dy_right, dz_down, dyaw]. Every waypoint is independently anchored at the current pose, not at the previous predicted waypoint. | | Normalization | dataset_statistics.json is authoritative. Actions use per-dimension q01/q99; padded action rows are zero after normalization. |

Benchmark adapters may declare a different action Protocol when required by the benchmark. The config and adapter Protocol are authoritative. See [Data Structure and State/Action Protocol](docs/guides/DATA_STRUCTURE.md).

Trajectory augmentation

Each animation aligns one raw trajectory with its enhanced version. Click an animation to open the MP4.

AerialVLN · Example 1

OpenFly · Episode 000008

Model

SimpleNav combines a vision-language backbone, selected long history, temporal-view context, and a continuous action head. The model consumes the protocol above and keeps dataset-specific coordinate semantics in the adapter.


Results

We adopt Qwen3.5-VL 4B as the unified vision-language backbone, and complete model training and closed-loop evaluation on 6 benchmarks respectively. Except for the necessary adaptation of data and task interfaces, we do not perform task-specific performance optimization for any individual benchmark. The results are summarized as follows. Full comparison tables and protocol notes are in [Release 01 Benchmarks](docs/guides/BENCHMARKS_RELEASE01.md).

| Benchmark | Split | NE↓ | SR↑ | OS/OSR↑ | SPL↑ | nDTW↑ | SDTW↑ | | --- | --- | ---: | ---: | ---: | ---: | ---: | ---: | | OpenFly | Seen | 37.1 m | 52.8 | 74.2 | 51.0 | - | - | | TravelUAV | Test Seen / Full | 85.6 m | 22.4 | 55.1 | 20.5 | - | - | | AerialVLN-S | Val Seen | 126.0 m | 8.4 | 18.9 | - | - | 3.4 | | R2R-CE | Val-Unseen | 4.7 m | 49.2 | 55.9 | 45.8 | - | - | | RxR-CE | Val-Unseen | 4.6 m | 58.4 | - | 52.2 | 74.6 | - |

| Benchmark | Task | SR↑ | TR↑ | CR↓ | | --- | --- |...