Meta's Digit 360 repository offers two software interfaces for the tactile sensor: a standalone Python path and a ROS 2 package. The Python example accesses sensor outputs directly; the ROS 2 interface publishes sensor modalities. This guide follows the public repository and its package documentation; we have not tested these instructions on Digit 360 hardware.

Build the documented ROS 2 workspace

The ROS 2 package guide starts with a virtual-environment script, then creates a digit360_ws/src workspace and clones the repository into it. From the workspace root, it invokes:

sh ./src/digit360/digit360/ros2/d360/colcon_build.sh

The guide next installs the Python package in editable mode with pip install -e . from the repository's inner digit360 directory. It then sources install/setup.bash from the workspace root. Those are the documented build and environment steps, rather than evidence that every host system is supported.

Choose raw or encoded images

The guide gives this minimal launch command for raw images:

ros2 launch d360 d360_min_launch.py image_pub_type:='raw'

For encoded images, substitute encoded for raw. The package documentation identifies encoded output as the default. It also provides a Foxglove visualization launch using d360_launch.py with raw images.

The minimal launch file makes that default explicit. Its raw branch chooses image_pub_opencv; the encoded branch chooses image_pub_encode. For each enumerated device, the file creates a namespaced group containing the image publisher, serial_io and audio_pub.

The same file checks that the device descriptor contains audio, image and data interfaces. Its assertion asks users to power-cycle a device when that descriptor is incomplete.

Launch individual streams

The ROS 2 guide lists separate commands for audio, temporal multimodal data and image output:

  • Audio: ros2 run d360 audio_pub
  • Temporal multimodal: ros2 run d360 sensors_pub
  • Image and visuotactile data: ros2 run d360 image_pub_opencv

These commands describe the repository's published software entry points. They are not a benchmark of sensing quality or robot-task performance.

Find the classification examples

A separate multimodal training guide documents a Conda environment, PyTorch installation and Foxglove data retrieval. Its examples train with RGB, IMU, microphone or spectrogram inputs, and show a combined run using all four flags. The data-generation command is python tools/gen_data_from_foxglove.py.

The repository's license file is labelled Attribution-NonCommercial 4.0 International. For the broader hand-level sensor platform, see our Meta Digit Plexus explainer.

Sources

This article was researched and fact-checked against the following sources: