Vitruvio: An Open-Source Leg Design Optimization Toolbox for Walking Robots
M. J. Chadwick, Hendrik Kolvenbach, Fabio Dubois, Hong Fai Lau, Marco Hutter
Abstract
We present an open-source framework for developing optimal leg designs for walking robots. The leg design parameters (e.g., link lengths, transmission ratios, and spring parameters) are optimized for a user-defined metric such as the minimization of energy consumption or actuator peak torque, enabling the user to better navigate through the high-dimensional and unintuitive design space. Our approach uses the single rigid body dynamics trajectory optimization tool TOWR to generate realistic motion plans. The planned predefined forces and motions are then used to identify actuator velocities and torques. Next, the leg design parameters are optimized using a genetic algorithm. The framework was validated by comparison with measured data on the ANYmal quadruped robot for a trotting motion, with errors in cumulative joint torque and mechanical energy each below 8% per gait cycle. Optimization of the ANYmal link lengths demonstrate that reductions in joint torque, mechanical energy, and mechanical cost of transport in the range of 5-10% are attainable.