One Python library for the real hand and its twin

The firmware can move motors. Now the computer needs a friendly way to tell it what to do.

One interface, two hands#

The new tendra Python package has one interface, Hand, with two versions behind it. SimHand drives the MuJoCo simulation, and RealHand talks to the ESP32 over USB. They share the same commands, so a script (or, later, an AI) written for one runs on the other.

from tendra import SimHand

hand = SimHand()
hand.set_joint("index_pip", 0.8)   # radians, positive = closing
hand.wait()                        # let the simulation settle
print(hand.positions())

Testing without hardware#

The hand still isn't ready to power on, so I also wrote FakeEsp32, a small piece of software that pretends to be the ESP32 and answers the same text commands. RealHand can't tell the difference, which means the whole chain from Python to the command language can be tested today.

There are 20 tests so far. Some check that the joint names and limits agree in three places: the firmware, the Python code and the simulation model. If one of them changes and the others don't, a test fails.

Digital twin v1#

A digital twin is a virtual copy of the hand that stays in step with the real one. In this first version, you move a slider in the MuJoCo viewer and the new target goes to the hand. To keep the USB link calm, it only sends when something changes, and at most 20 times a second.

uv run python sim/twin.py --fake

With --fake, the software ESP32 stands in for the hand. In a first test, the viewer, the simulation and the link all ran together and shut down cleanly.

Something to tune#

At rest, the simulated thumb sags about 1.2° under its own weight, because its simulated motor is a little soft. A real stepper holds firmly. That's a setting to tune once I can compare with the real hand.

Next: power on the real hand, check every motor's direction, and build a guided tool to calibrate each joint.