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.

Related Work: ObjectReact [CoRL'25] | RoboHop [ICRA'24]

Results

<div align="center"> <div style="max-width: 800px;"> <table align="center" width="80%"> <caption> Navigation success rate across varying trajectory lengths. </caption> <thead> <tr> <th>Controller</th> <th>Easy [1–3m]</th> <th>Hard [3–5m]</th> <th>Full [8–10m]</th> </tr> </thead> <tbody>  <tr> <td colspan="4" align="center"><strong>GT‑Metric</strong></td> </tr> <tr> <td>RoboHop[1]</td> <td align="right">93.14</td> <td align="right">78.43</td> <td align="right">42.16</td> </tr> <tr> <td>PixNav[2]</td> <td align="right">65.69</td> <td align="right">44.12</td> <td align="right">15.69</td> </tr> <tr> <td><strong>Ours (TANGO)</strong></td> <td align="right"><strong>94.12</strong></td> <td align="right"><strong>90.20</strong></td> <td align="right"><strong>48.04</strong></td> </tr>  <tr> <td colspan="4" align="center"><strong>GT‑Topological</strong></td> </tr> <tr> <td>RoboHop[1]</td> <td align="right">78.43</td> <td align="right">58.82</td> <td align="right">25.49</td> </tr> <tr> <td>PixNav[2]</td> <td align="right">60.78</td> <td align="right">44.12</td> <td align="right">15.69</td> </tr> <tr> <td><strong>Ours (TANGO)</strong></td> <td align="right">74.51</td> <td align="right"><strong>65.69</strong></td> <td align="right"><strong>30.39</strong></td> </tr>  <tr> <td colspan="4" align="center"><strong>No‑GT</strong></td> </tr> <tr> <td>RoboHop[1]</td> <td align="right">43.56</td> <td align="right">34.56</td> <td align="right">13.73</td> </tr> <tr> <td>PixNav[2]</td> <td align="right">51.96</td> <td align="right">39.22</td> <td align="right">14.00</td> </tr> <tr> <td><strong>Ours (TANGO)</strong></td> <td align="right"><strong>61.76</strong></td> <td align="right"><strong>43.14</strong></td> <td align="right"><strong>21.57</strong></td> </tr> </tbody> </table> </div> </div>

[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/

Docker Setup (Alternative)

Prerequisites

Docker >= 20.10.0 with NVIDIA Container Toolkit installed (installation guide)
Docker Compose >= 2.0.0
NVIDIA GPU drivers

Quick Start

Set up X11 display forwarding:

xhost +local:docker
touch /tmp/.docker.xauth
chmod 666 /tmp/.docker.xauth

Build and run the container:
```
docker-compose up --build
```

Volume Mappings

The following directories are mapped from host to container:

./logs → /app/logs - Application logs and debug output
./outputs → /app/outputs - Visualization outputs, videos, plots
./data → /app/data - Input data and datasets
./third_party/models → /app/third_party/models - Model weights
./configs → /app/configs - Configuration files

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's if you find them helpful :)

TANGO

@INPROCEEDINGS{11127998,
  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)}, 
  title={TANGO: Traversability-Aware Navigation with Local Metric Control for Topological Goals}, 
  year={2025},
  volume={},
  number={},
  pages={2399-2406},
  keywords={Measurement;Visualization;Image segmentation;Three-dimensional displays;Navigation;Source coding;Pipelines;Trajectory;Planning;Robots},
  doi={10.1109/ICRA55743.2025.11127998}
}

ObjectReact

@inproceedings{podgorski2025tango,
  title={TANGO: Traversability-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)},
  pages={2399--2406},
  year={2025},
  organization={IEEE}
}

Robohop

@inproceedings{garg2024robohop,
  title={Robohop: Segment-based topological map representation for open-world visual navigation},
  author={Garg, Sourav and Rana, Krishan and Hosseinzadeh, Mehdi and Mares, Lachlan and S{\"u}nderhauf, Niko and Dayoub, Feras and Reid, Ian},
  booktitle={2024 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={4090--4097},
  year={2024},
  organization={IEEE}
}