Educational robotics · Human–Robot Interaction
An educational robot and a learning platform built as one system, so children can meet robotics, AI, and programming through something that looks back at them.
AICOS pairs an embedded robot — Raspberry Pi and Arduino, camera, microphone, speakers, LED face, and a touchscreen — with a distributed, event-driven platform running a conversational assistant, a recommendation engine, and real-time services. A no-code block editor sits between the two, so a child can program movement, expression, and dialogue without writing a line of code.
Robot architecture
A pixel face carries every emotion the robot needs: eyes and mouth are drawn on an LED matrix behind a smoked diffuser, so expressions stay readable across a noisy classroom. Camera apertures sit above and below the panel, and the whole head turns on a servo neck, so attention is something a child can actually see.
An 8×8 RGB matrix beats on the chest as a heartbeat — idle, listening, thinking, and speaking each get their own pattern, so the robot never looks frozen while it works. The touchscreen beside it carries the block editor and anything easier tapped than spoken.
Two processors divide the work by what each is good at. A Raspberry Pi runs Linux and everything that thinks: speech, vision, behavior scripts, and the link to the platform. An Arduino owns everything that must never stutter — motor timing, LED refresh, sensor polling — and the two talk over a serial link.
A laser-cut MDF chassis, slot-and-tab assembled, keeps the robot repairable by the people who built it: every panel comes off with a screwdriver. Motor drivers sit under the deck with the battery pack, whose voltage is polled continuously so a low charge dims the face instead of cutting a session mid-sentence.
A child talks; the robot answers out loud. Transcription and synthesis run on the robot wherever they can, so an ordinary question never has to leave the room.
Vision drives attention rather than surveillance: the robot notices someone is there, turns toward them, and reacts on the LED face. Frames are processed locally and not stored.
Blocks snapped together in the editor compile into a behavior script, published over MQTT, and executed by the microcontroller — a child sees their program move the robot within seconds.
Learning platform
The platform is where the robot stops being a toy and becomes a curriculum. Services are split by responsibility and speak whichever protocol fits the job — gRPC where latency matters, GraphQL where the client shapes the query, REST at the edges, and Kafka whenever one service’s result is another service’s trigger.
4000500514001400250052400340044005The public repository covers the platform side only — the robot firmware and the embedded stack are not part of it.
A Scratch-inspired block editor turns movement, animation, expression, and dialogue into pieces children can snap together. Robotics becomes something they author, not something they watch.
Speech and vision run on the robot itself. Keeping perception at the edge means a classroom can use the robot without a child’s voice or face becoming someone else’s dataset.
Face, heartbeat, head turn, and voice are designed as one signal set, so the robot’s state is legible at a glance — attention, thinking, and answering never have to be explained.
Recognition
