End-effector payload inertial parameter identification for industrial robots via co-trajectory differenced reduced-order observation and manifold optimization

Accurate identification of end-effector payload inertial parameters is fundamental to precision control of industrial robots. Space and collision-avoidance constraints make full-stroke excitation difficult in practice; conventional identification also faces four coupled difficulties: base-joint nonlinear coupling, rank deficiency under equation reduction, non-Gaussian outliers, and physical-feasibility violations. To address these four difficulties, this paper proposes an end-effector payload identification method based on co-trajectory differenced reduced-order observation and Riemannian manifold optimization. The method uses co-trajectory differenced reduced-order observation on joints J3–J6 to avoid proximal nonlinearities, and combines Tukey and LAD weighting to suppress outliers; a subsequent two-stage Riemannian manifold optimization then fuses redundant measurements within the SPD pseudo-inertia domain and strictly enforces physical feasibility via spectral truncation. Comparative experiments on a six-axis industrial robot show mass identification errors of 0.08 % and 4.7 % under full and medium load, and about 12.4 % under low torque-ratio light load, remaining substantially better than the baselines, while the identified parameters stay strictly inside the pseudo-inertia cone under all loading conditions.