The starting idea here is charming: a lamp built as a companion for the home office, one that can even pose as a fake lamp running on a laptop. What began as a playful concept has turned into a serious piece of robotics work, and the pace of development in October is remarkable. The main branch logged 4,092 commits that month, blowing past the previous record of 1,615 set back in April, all since the first commit landed on 24 March.
The hardware looks tidy and well thought out. An articulated arm in beige segmented plastic moves on motorized joints, sitting on a square base with vents. Up top, the head pairs a conical lampshade with an LED ring light around a central dark camera aperture. STS3215 servos provide position feedback, and the Fusion 360 file for the current body design is available, which anyone considering a similar build should appreciate.
The behavior is where this shines. The lamp ties every line of dialogue to the face and voice that produced it, and keeps each person's memory separate. Position feedback lets it tell a friendly push apart from a fault, and after being pushed it stops what it is doing and holds still for 30 seconds. Treat a joint pulled away from its goal as a hand, and the intervention detection practically writes itself. If an arm strains three times, the lamp sits down and rests for 30 minutes. Audio latency is impressive too: the first part of an answer starts coming out within 400 ms, and the forgetting function dropped from 9 seconds to under 400 ms as well.
There is still room to grow. Mood reading sits at 39% on the internal tracker, and recognizing two people in the same room is built but switched off by default. With Python and JavaScript SDKs, lampOS at 0.0.4 and 0.0.5, and a documented path in docs/develop.md, the project seems set up for others to dig in.