Research Semi-Direct Drive Walking with TipTap 3D-Printed Bipedal Robot

Aerospace engineer Darren Levine wanted to learn the low-levels of math and physics behind control algorithms needed for semi-direct…

Aerospace engineer Darren Levine wanted to learn the low-levels of math and physics behind control algorithms needed for semi-direct drives that power some bipedal robots, and what better way to do that than by building his own 3D-printed desktop version. Levine began working on his TipTap robot two years ago, and at this point in its development, it looks like hard work is paying off with his latest design.

TipTap was designed as a model for testing control algorithms and simulation fidelity utilizing a semi-direct drive. (📷: Darren V Levine)

On the mechanical end of the design spectrum, Levine went with a naturally unstable pointed foot model that requires constant control to remain stable, which is where the semi-direct drive comes into play, as it offers enough torque control to stay upright. His drive features four 3D-printed 8:1 planetary gearboxes that are back drivable and provide a metal ball bearing load path.

The TipTap robot is also outfitted with four direct-servo controllers that handle position feedback and torque control, 4DOF legs with GBM3506H-130T gimbal motors, 2DOF hips with MG92B servos, and low-limb inertia belts that isolate the drive extremities from the drive system. Controlling the show is a BeagleBone Blue SBC that provides wireless access and control, as well as a 9-axis IMU and 6V servo rail for managing all those motors. You can see how the legs function in the GIF below.

Most parts of the TipTap robot — minus the electronics and servos — were 3D-printed (using PLA) and are quite durable for its funky dance applications. As for software, Levine developed a Port-Hamiltonian simulation framework in MATLAB, to get up to speed on the low-level math and physics required for the necessary algorithms and simulation research. He has since moved on to an articulated body with Featherstone simulation framework, again using MATLAB.

“Eventually I’ll work on efficiency and stability, but for now this does have the advantage of being naively extensible to arbitrary N-Jointed x N-Limbs without a training period. My next steps are to progress more on the hardware end in order to test any controllers on actual hardware.

Those interested can follow his TipTap progress on his project page.

Latest Articles