Skip to main content

AnyTraverse

An offroad traversability framework with a VLM and a human operator in the loop.

arXiv pypi CI GitHub Repo stars

News 📰

  • Sep 2026: Released v2 with improved API, more VLMs, and benchmarking data.
  • Jul 2025: AnyTraverse presented at the 19th International Symposium on Experimental Robotics, Santa Fe, New Mexico. Set to be published in proceedings (Springer Nature, "Experimental Robotics").

Installation

Requires Python 3.12 or newer.

  1. Install PyTorch first, so you get the right build for your platform (CUDA, MPS, CPU, or a Jetson wheel).
    uv pip install torch       # uv users
    pip install torch          # pip users
    

    ⚠️ PyTorch does not provide wheels for the NVIDIA Jetson platform. On Jetson, install the matching torch wheel first: JetPack 6 via https://pypi.jetson-ai-lab.io/jp6/cu126, JetPack 7 via https://pypi.jetson-ai-lab.io/sbsa/cu132. The hf extra below then pulls a matching torchvision.

  2. Install AnyTraverse with the model backends you need.
    uv pip install "anytraverse[hf]"       # uv users (Hugging Face VLMs)
    pip install "anytraverse[hf]"          # pip users (Hugging Face VLMs)
    pip install anytraverse                # core only, for custom models
    pip install "anytraverse[viz]"         # plotting helpers for the quickstart tutorial
    
    SAM 3 weights are gated: accept the license for facebook/sam3 and run hf auth login once.

Docs

  • Quickstart 🏃‍♂️ — build the paper pipeline, run one step, and plot attention, traversability and uncertainty maps. Also covers swapping VLMs/encoders (build_pipeline_sam3, build_pipeline_grounded_sam2), the operator-in-the-loop calls, and deployment notes (dtypes, caching, state.to("cpu") for ROS).
  • Benchmarks 📊 — measured latency and VRAM per model on an RTX A4500, end-to-end pipeline estimates (the paper pipeline runs in ~8 ms in ~0.6 GB), and a step-time vs. prompt-count plot for 1–10 prompts.

Launch AnyTraverse with ROS 2 🤖

AnyTraverse integrates with the Nav2 stack in ROS 2. See sattwik-sahu/anytraverse_ros to learn more.


Breaking changes in 2.0

  • Flat package layout: from anytraverse import AnyTraverse, CLIPSegAttentionMapping instead of anytraverse.utils.*.
  • Spelling fixes: TraversalState.UNKNOWN_OBJECT (was UNKOWN_OBJ), init_traversability_preferences (was init_traversabilty_preferences).
  • The Nomic encoder was removed; use siglip2 (default in the new presets) or dinov2.
  • requires-python = ">=3.12", torch >= 2.4; model weights now use the standard Hugging Face cache.

Contributing 👨‍💻

We'd love to see your models and improvements. Please open a pull request (branch name: dev/feat/<your-feature-name>) and an issue to discuss a new feature. Run ruff check ., ruff format --check . and pytest before pushing.


Made with ❤️ in IISER Bhopal.

Metadata

Release files for anytraverse 2.0.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for anytraverse 2.0.0
File Size Uploaded Provenance Actions
anytraverse-2.0.0.tar.gz 1.7 MB Without provenance Details for anytraverse-2.0.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for anytraverse 2.0.0
File Interpreter ABI Platform Actions
anytraverse-2.0.0-py3-none-any.whl Python 3 none any Details for anytraverse-2.0.0-py3-none-any.whl

Total release size: 1.7 MB

Release files / anytraverse-2.0.0.tar.gz

Download URL anytraverse-2.0.0.tar.gz
Size 1.7 MB
Tags Source
SHA-256 checksum
How to use checksums
eddd12624f155a84bf184c70102f13dcc770b9d0615698acf8f4d3ebac1f83f4
BLAKE2b-256 checksum
How to use checksums
adce95f7ff562d2e5b251ebe22a9222cda97a311da1e051c3bf83aaa5941bc4a
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.14 {"installer":{"name":"uv","version":"0.12.14","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Release files / anytraverse-2.0.0-py3-none-any.whl

Download URL anytraverse-2.0.0-py3-none-any.whl
Size 57.5 kB
Tags Python 3
SHA-256 checksum
How to use checksums
fd0b23f85de311db1a3c36dac04a7a77ce7cc4b5a8d7b782738bf3b39de1903c
BLAKE2b-256 checksum
How to use checksums
11f566dea7104edab74d8b692709db9596a8fb20158224caa6c54ee05961a3e6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via uv/0.12.14 {"installer":{"name":"uv","version":"0.12.14","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

Security for anytraverse 2.0.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page