Task-Aware Actuator Parameter Allocation for Multibody Robots

Abstract

Legged robots demand actuator parameter sets that reconcile conflicting requirements such as high torque density, low impedance, responsiveness, and energy efficiency. This paper presents a taskaware codesign framework that jointly optimizes actuator parameters and motion for humanoids. An outer loop searches the continuous space of motor mass and gear ratio using CMAES; datadriven regressions fitted to manufacturer catalogs recover peak torque, rotor inertia, motor constant, geometry, and stagedependent efficiency for each joint. An inner loop performs fullbody constrained trajectory optimization with torque–velocity, kinematic, and contact constraints, and evaluates energy via an electrical model that separates Joule/parasitic losses and gearbox friction. An ablation study on these loss terms isolates their impact on actuator allocation. We study three pelvis topologies across scenarios including loaded/unloaded walking, a forward jump, stair ascent with a 10 kg payload, and a manipulation “workout” with handheld weights. Simtosim validation with an RL locomotion policy (IsaacLab to MuJoCo) confirms feasibility and yields coherent ranges of electrical energy and costoftransport. The framework returns jointspecific motor masses and gear ratios that expose interpretable tradeoffs, providing a practical recipe for topologyaware actuator allocation in humanoids. GitHub repository: <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/SmbdOnceTldMe/actorob</uri>.