I don't do CAD anymore. Every part is a Python program that builds the model in memory using build123d, an open-source parametric modeling Python framework. It then exports the model as a STEP file, which Fusion 360 imports so they can be inspected visually. A high-level YAML spec file holds the numbers that define the robot as a whole - dimensions, masses, actuators, joint limits - which is read by the code. The python code obviously is written by an LLM (I currently use Opus). I did consider using Onshape, but its API is capped to a few thousands calls a year, which make it unusable for me: I use Opus to push the parts, build the assembly, create the joints, set their limits and values, assign materials and densities, and inspect the result. That's easily a few hundred calls a day.

This is Q2, a redesign of my Q1 quadruped. A major simplification and an AI-first design: I first made sure the design reaches fast RL convergence in simulation, and only then moved on to CAD. - Four identical legs: no left/right parts - Direct-drive knees: no linkage or belts - No reduction gear after the motors output - Knees rotate 360°: no predefined knee posture - Low standing load: almost no knee torque needed - Shorter links: so lower leg inertia and joint torques - Battery at the geometric center, so a nicer CoG - One spec file drives both the sim model and the CAD, so CAD is essentially written in code. This is parametric CAD taken to the extreme: the CAD software is only used to review the design, not to build it Weight, cost, manufacturing time and assembly effort would also drop significantly. Same battery design, same PCB, same compute.