All the code we produce for our robots is open source, developed in the open on our GitLab instance and released to the community as installable packages through our APT repository.

The stack is broad. Thematically it covers perception (face and body detection, person tracking), speech (ASR and TTS), human-robot interaction skills and message interfaces, simulation, and the infrastructure (CI/CD pipeline, architecture-level tooling) needed to build and ship complete robot applications. In terms of ROS 2 support, we target the Humble, Jazzy and Lyrical distributions, each built for both amd64 and arm64 architectures.

Some highlights:

• ROS4HRI: we are among the main developers of ROS4HRI — the standard ROS API for human-robot interaction — and we maintain a large number of conformant packages around it. See the ROS4HRI project page for details.

  

• Sound processing: we develop and package the audio stack end to end — microphone front-ends, Whisper-based automatic speech recognition, and text-to-speech engines (e.g. Chatterbox, CosyVoice 3) — shipped as ready-to- install APT packages.

• reMap: reMap is a framework for dense 3D world representation with a plugin architecture for faces, objects and navigation maps, primarily developed by Lorenzo Ferrini.

• Architecture-level tooling: schemas, linters and introspection tools to validate the organization of our software components (missions, tasks and skills).

• Dialogue management: our dialogue manager supports multi-user, multi-turn conversations between robots and several people simultaneously.