MiniBot

A compact desktop robot built to explore how a small physical machine can host a full local-AI stack — voice, vision, language, and motion — without depending on the cloud.

Problem

Most “AI robot” demos rely on a remote LLM, cloud vision APIs, and a tethered laptop. The interesting questions — latency, privacy, on-device model selection, full autonomy when the network drops — get hidden behind someone else’s servers.

MiniBot asks: how much can a Raspberry Pi–class machine actually do on its own?

Concept

A small, expressive robot body that runs entirely on local compute. It listens, looks, thinks and moves. It can answer questions, describe what it sees, follow a face, and react to what’s in front of it.

What it demonstrates

• On-device speech recognition and synthesis
• On-device vision pipelines (face detection, object detection)
• Local language model integration without cloud round-trip
• Real-time motion control for expressive behaviour
• System orchestration across sensors, models and actuators

Technology

• Raspberry Pi compute
• Local AI / on-device LLM inference
• Voice input + output
• Computer vision
• Servo-driven motion
• ROS 2 (experimenting)

Status

PROTOTYPE — software stack is being iterated. Body and motion platform are working. Next: better on-device vision and a more stable voice pipeline.

Updates

• First successful end-to-end “see + think + speak” loop on-device.
• Servo calibration routine landed.
• Voice pipeline swapped to a fully local recogniser.