Dynamic Parameter Identification of a Hybrid Bipedal Robotic Leg via Current-Offset Compensation and Trajectory Optimization

Parameter identification for hybrid bipedal robotic legs remains challenging because parallel mechanisms introduce strong dynamic coupling and nonlinear friction is difficult to model accurately. This paper presents an enhanced identification method for a six-degree-of-freedom hybrid robotic leg with a 3-DOF serial hip, a serial knee, and a 2-DOF parallel ankle. The closed-loop kinematics of the parallel ankle are formulated, and a motor-to-joint torque mapping is derived using the principle of virtual work. The hybrid mechanism is then transformed into an equivalent serial multibody system for Lagrangian dynamic modeling. To improve model accuracy, motor current offsets are incorporated into the identification process to compensate for zero drift in low-torque regions. In addition, a trajectory optimization criterion formulated as a mass-weighted sum of the condition numbers of link-wise sub-regressor matrices is introduced to improve the balance of parameter excitation, subject to nonsingularity and excitation constraints. Experiments on the physical robotic leg show that current-offset compensation provides modest improvements under the baseline objective, while the final configuration combining the unified objective with current-offset compensation achieves the best overall performance. Compared with the baseline configuration, the final configuration reduces the NRMSE of Joint 1 and Joint 6 by 53.3% and 40.1%, respectively.