Easter Special

[Nathaniel Nifong]

(New Video) Robot hunts for Easter eggs

youtube.com/watch?v=hb1P9MxVfys

As a challenge I decided to test Stringman's abilities by training it to hunt for Easter eggs in my room. This requires all the systems to be working well together and to collect a lot of high quality data to train on. It required my uninterrupted attention for three days straight, but in the process I ironed out several rough spots and made Stringman easier to train than before. If you've ever wanted to learn about imitation learning in robotics, this is a great platform to try it on since you can reach more than just a desk and can train it to do a wide variety of tasks in a room.

PCBs ready for assembling new Anchors

Circuit boards for the new Stringman anchors came in and I've been assembling the first units. (the initial prototype of these anchors are the ones I trained the Easter egg hunting AI on). The passive corners have an aconite (ceramic) ring from a fishing rod to serve as a pull point. So now only two corners require power outlets and the other two are barely visible. The full build guide for these can be found at https://neufangled.com/docs/arpeggio_anchor_build_guide/

Train your own models without AI experience

When a fellow robotics founder asked me once if it was easy to train Stringman to do new tasks, I had to be honest and admit that it was pretty hard. That's been nagging at me ever since, so I made it easy. Now you can just click record in the UI and drive it around to record a dataset. Click a button in a Google Colab script to train the model with cloud credits, then another button in the UI to try out your trained model the next morning. I think it's fair to say that it's not "hard" anymore but if you want to prove me wrong, be my guest.