TANGO: Traversability-Aware Navigation with Local Metric Control for Topological Goals [ICRA 2025]

This is the official implementation of TANGO. [PDF] [Project Page]

This repository contains the TANGO controller, and supports testing the gt_metric source of topological goals obtained from the simulator. For complete evaluation, please refer to this.

Results

Navigation success rate across varying trajectory lengths.

Controller	Easy [1–3m]	Hard [3–5m]	Full [8–10m]
GT‑Metric
RoboHop[1]	93.14	78.43	42.16
PixNav[2]	65.69	44.12	15.69
Ours (TANGO)	94.12	90.20	48.04
GT‑Topological
RoboHop[1]	78.43	58.82	25.49
PixNav[2]	60.78	44.12	15.69
Ours (TANGO)	74.51	65.69	30.39
No‑GT
RoboHop[1]	43.56	34.56	13.73
PixNav[2]	51.96	39.22	14.00
Ours (TANGO)	61.76	43.14	21.57

[1] RoboHop-Garg S. et al., Robohop: Segment-based topological map representation for open-world visual navigation ICRA 2024.

[2] PixNav-Cai et al., Bridging zero-shot object navigation and foundation models through pixel-guided navigation skill, CVPR 2024

Installation

Clone this repo - recursive is required for Depth Anything

git clone --recursive https://github.com/podgorki/TANGO.git
cd TANGO

Create new environment

python3.10 -m venv .venv --prompt tango
source .venv/bin/activate
python -m pip install --upgrade pip setuptools wheel

Install controller and sim (required for demo)

Pre-install habitat-sim

Dependencies

sudo apt-get install -y --no-install-recommends libjpeg-dev libglm-dev libgl1-mesa-glx libegl1-mesa-dev mesa-utils xorg-dev freeglut3-dev
pip install cmake==3.14.4
pip install "numpy>=1.25,<2" --upgrade  # required before building habitat-sim

Clone and build the Sim (takes a bit)

cd third-party/
git clone https://github.com/facebookresearch/habitat-sim.git
cd habitat-sim/
git checkout v0.2.4
python setup.py install --cmake
cd ../..

Install TANGO

pip install -e ".[habitat-lab]" --extra-index-url https://download.pytorch.org/whl/cu128 --prefer-binary

Depth anything

Depth anything is installed by submoduling.

Add a pth so you can resolve zoedepth

echo "$PWD/third_party/depth_anything/metric_depth" > \
     $(python -c "import site, sys; print(site.getsitepackages()[0])")/zoedepth_local.pth

The depth anything model weights are located at: https://huggingface.co/spaces/LiheYoung/Depth-Anything/tree/main/checkpoints_metric_depth And also grab the vit from here https://huggingface.co/spaces/LiheYoung/Depth-Anything/tree/main/checkpoints place them in third_party/models/

TANGO Demo

Download official hm3d v0.2 following their instructions. Place (or synlink) it at ./data/.

Unzip test data: unzip data/hm3d_iin_val.zip -d ./data/

python -m scripts.run_goal_control_demo

BibTex

Please cite our paper if you find it helpful :)

@inproceedings{podgorski2025tango,
  title={TANGO: Traversablility-Aware Navigation with Local Metric Control for Topological Goals},
  author={Podgorski, Stefan and Garg, Sourav and Hosseinzadeh, Mehdi and Mares, Lachlan and Dayoub, Feras and Reid, Ian},
  booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)},
  year={2025},
  organization={IEEE}
}