Getting started

Tendra Hand is an open-source robotic hand that is 3D-printed and driven by tendons, like a human hand. The long-term goal is a robot that can do what people do, like cooking and chores, as well as a person, and the hand is where it starts: a hand with the same joints as yours. Everything is open: the CAD files, firmware, Python code, simulation and research notes.

Right now the project is in Phase 1: a thumb and an index finger with 8 joints, each driven by its own motor. You don't need the hardware to start. The simulation runs on any normal laptop.

What's in the repository#

What you need#

• Git, to download the code.
• uv, a fast Python package manager. It installs the right Python version (3.12) for you, so you don't need to set up Python yourself.
• Any recent Windows, macOS or Linux computer. No graphics card is needed: MuJoCo runs on the CPU.

Get the code#

git clone https://github.com/Danielloka/tendra-hand.git
cd tendra-hand

Install the Python environment#

From the repository root:

uv sync

This creates a private Python 3.12 environment in .venv/ and installs MuJoCo, NumPy, pySerial and the tendra package itself. You only need to do it once (and again after pulling changes that add dependencies).

Open the hand in the simulator#

uv run python sim/view.py

A MuJoCo window opens with the thumb and index finger. In the right-hand panel, open Control: there is one slider per motor (M1 to M8). Drag them to bend the joints.

Try the digital twin without hardware#

A digital twin is a simulated copy of the real hand that stays in sync with it. You move the sliders in the simulator, and the same targets are sent to the real hand. With --fake, a software stand-in for the ESP32 takes the place of the real board, so you can try it with no hardware at all:

uv run python sim/twin.py --fake

The terminal shows how far the (fake) hand is from the simulation as it catches up.

Use the Python API#

The same Hand interface drives both the simulation and the real hand, so code written for one runs on the other. Angles are in radians, and positive always means closing the hand.

from tendra import SimHand

hand = SimHand()                     # the MuJoCo simulation
hand.set_joint("index_pip", 0.8)     # bend the index middle joint ~46°
hand.set_targets([0.5] * 8)          # all joints, motor order M1..M8
hand.wait()                          # run the physics until the joints arrive
print(hand.positions())              # measured angles, in radians

Swap SimHand() for RealHand() to talk to the real hand over USB. See software/README.md for the full list of methods.

Run the tests#

uv run pytest

The tests check the simulation model and the Python package. They also make sure the joint names and limits agree across the firmware, the Python code and the MuJoCo model, so a change in one place can't silently break another.

Where to go next#