Teo
Brown

ABENICS is a joint mechanism that mimics a ball joint. Two monopole gears drive a cross spherical gear, so the joint turns about all three axes. My research uses it as a robot wrist. I wrote a SolidWorks script that generates both gears from the ABENICS paper, and printed a PLA set to confirm that two monopole gears mesh with one cross gear.

To get the load, I put a UR5e arm with an AmazingHand in MuJoCo and ran a full pick-and-place cycle with a torque sensor at the wrist. The static hand calc gives 0.0886 N-m. Motion adds 7.7 percent, so the 0.0954 N-m peak became the design load.

Then single-tooth FEA: the motor gear pushes one tooth until each of four printed materials reaches its strength, checked along and across the printed layers. The weakest case, PETG HF across its layers, still holds 6.8 times the peak load.

Next: tensile bars printed on the lab's printer, then the FEA again with measured properties. A SolidWorks rig that loads 20 gear sets to failure is in design.

Standard FEA treats a printed part as one solid material. A real print is stronger along its printed lines than across its layers, and its infill is mostly air. I proposed building the model straight from the sliced G-code. Every element gets its layer, its line directions and how full it is. CalculiX solves it, and each line direction is checked for breaking within a layer or between layers.

On a 24-tooth spur gear printed standing on its teeth, the tooth pointing up splits between layers at about 12.4 N. Standard FEA says 15.5 N, and it can't tell that tooth from one on the side, which cracks through a layer instead.

It passes 44 of 44 verification cases against hand calcs, with the limits written down before the first run. Those checks caught a solver bug: CalculiX's multithreaded solver gave wrong answers in 10 of 24 repeat runs, with no error. Every run now uses a different solver and rejects any solve whose forces don't balance.

The material constants are datasheet placeholders until my coupon tests, so these forces test the method, not the parts.

A micromouse is a small robot that solves a maze on its own. I designed our drive module: two Faulhaber 1717 DC motors share one printed mount, and a 16-tooth pinion on each motor drives two 60-tooth spur gears, one per wheel.

I picked the drivetrain by testing gear contact ratio, gear material and heat transfer between materials. I designed more than five of the robot's main parts in Fusion 360 and worked with the electrical team on the PCB in KiCad.

The skid plate under the PCB took the most iteration. After three printed versions of a friction slider, I replaced it with a rolling ball.

GELLO is UC Berkeley's open-source teleoperation controller: a scaled copy of the robot arm made from printed parts and DYNAMIXEL servos. Move it by hand and the real arm follows.

I printed and assembled one for the lab's UR5 and designed its servo mounts and cable routing. I set the seven servo IDs 1 to 7 from base to gripper, then calibrated each joint's offset so the same pose on the GELLO means the same pose on the robot.