Thor: an industrial joint layout for EUR 350, paid for in printing hours
A 6-axis arm built from stepper motors driving printed gears and GT2 belts, with a joint arrangement copied from industrial manipulators. The trade is typical of the genre: the money you save is converted directly into printer time.
- Structural parts
- 49 STL files, all designed in FreeCAD
- Print time
- Estimated 60-80 hours at 0.4 mm nozzle, 0.2 mm layers
- Specs
- 625 mm tall, 750 g payload, 6 axes plus gripper
- Parts cost
- About EUR 350 (printer and tools excluded)
- Transmission
- Stepper motors with printed gears and GT2 belts
- Firmware and software
- GRBL (accepts G-code); Asgard host app; ROS 2 / MoveIt2 integration
- Licence
- CC BY-SA 4.0
- Files you actually get
- 49 STL parts cover the six joints, the gripper, the base and bearing fixtures, all designed in FreeCAD. The control board is a custom shield on an Arduino Mega that combines six stepper drivers with limit-switch and fan inputs. The firmware is GRBL-based, meaning it accepts standard G-code — and that matters more than the parts list, because it puts the arm inside a tooling ecosystem that has existed for over a decade.
- BOM and delivered cost
- Parts come to about EUR 350. The dominant cost is time, not money: 60 to 80 hours of printing is the baseline, and failed reprints scale on top of that. A commercial 6-axis arm is in a different price universe, but you hold every structural file here, so any part that cracks can be printed again.
- Printing and machining
- 60-80 hours assumes everything works first time. Gear and belt transmission demands more from a printer than cosmetic shells do: layer bonding and tooth accuracy feed directly into joint backlash, and both depend on nozzle, layer height and filament. If you are new to tuning, print two gear pairs in scrap material and measure the backlash before committing the whole set.
- The hard step
- Not assembly — getting backlash across six joints small enough for the kinematics to converge. Printed gears and belts have backlash by nature, so a target position approached from two directions ends up in two different places. Every printed arm faces this; the answers are preload, compensation, or better transmission, and all three cost time.
- What we checked
- We checked: the STL manifest against the joint units, the firmware's command protocol (whether it is standard G-code), and how the ROS 2 / MoveIt2 integration is packaged. Not checked: we have not assembled this arm, so the 750 g payload and repeatability figures rest on vendor and community claims, not on our own measurement.
If you want to learn kinematics and own a multi-axis arm you can repair and modify, the choice here is straightforward — accept that the real cost is printer time. It is wrong for anyone who wants something moving within a week, and right for someone who will spend a quarter building one machine and keep revisiting it afterwards.
Why the G-code detail is worth money
Many homebrew arms ship proprietary firmware, so connecting anything else means writing a bridge yourself. Thor accepts G-code directly, which plugs it into tooling that already exists: trajectories can be generated by off-the-shelf software, recorded G-code can be replayed as a motion, and the ROS 2 side has a bridge already written.
60-80 hours assumes success on the first pass. In practice the parts scrapped while dialling in settings tend to consume around a third of the material again — buy 1.5x the filament rather than stopping half-built when you run out.
Sources
- Project repository (STL manifest, hardware notes, licence)
- Official documentation and the accompanying ROS 2 / MoveIt2 repositories
Objects in this entry
Last checked 2026-09-28