For rapid pick-and-place operations using parallel mechanisms, separate trajectory planning and control can lead to mismatched constraints, overly conservative safety margins, and inefficient motor utilization. This study proposes an integrated framework in which both the planner and the controller use the same simplified rigid-body dynamic model. A dynamics-constrained time-optimal planner computes the optimal velocity profile offline and provides online reference values through interpolation at 1 kHz. Computed torque feedforward is combined with a generalized momentum disturbance observer that avoids online inversion of the inertia matrix. The framework is deployed on an ARM and EtherCAT embedded platform and validated through processor-in-the-loop (PIL), in which the controller executes all online algorithms in real time on the embedded hardware while the mechanical system is replaced by a high-fidelity virtual prototype that incorporates distributed rod inertia, joint friction, transmission elasticity, and parameter perturbations absent from the design model. For a standard gate-shaped trajectory, the cycle time decreases from 0.475 s to 0.392 s, a 17.5% reduction, compared with S-curve planning with PD control. The peak tracking error increases by only 0.10 mm or 3.3% (3.16 mm vs. 3.06 mm over 50 runs per group), while RMS error decreases by 34.1% to 1.18 mm. Group B (time-optimal planning with PD control only) shows a maximum error increase of 86.3%. The worst-case control-loop execution time is 121 μs, accounting for 12.1% of the cycle period.
