The hardest part here is clearly the locomotion control itself, and the maker spent the last few weeks debugging and fine-tuning the controller to get it where it stands now. The framework combines MPC with whole-body control, and a depth camera builds a height map of the ground ahead. That pairing seems to be paying off, since stairs are now climbed more reliably than before.
The robot itself has motorized actuators at every leg joint plus rubberized foot pads, and there is 18 cm of ground clearance, enough to step over obstacles up to 11 cm tall. The white, red and black chassis carries an NDPC label, and some internal wiring is visible, which I find refreshing on a research build. Everything is written in C++ and works with ROS2, and the whole controller is open source on GitHub, a choice I think is worth a lot for anyone wanting to study this kind of system.
The maker ran three different obstacle courses. In one, the robot climbs yellow foam steps up to a raised platform. In another, it goes up a low barrier of stacked flat panels and comes back down. In the final one, it walks across an irregular pile of loose materials and keeps its footing the whole way. Soft surfaces are also handled fairly confidently now.
One open question is which actuators are used and what the total build cost is, since that has not been answered yet. The next goals are on the software side: automatic device connectivity tests for reliability, and standing up from random positions.