Physics-based computer simulations are a powerful tool to define personalized robotic rehabilitation treatment planning for individuals with spinal cord injury (SCI). Current models often lack detailed representation of upper body muscle forces, which are crucial as these subjects often rely on walkers for mobility. This work investigates if adding physiological muscle torque actuators improve the prediction of upper body mechanics compared to having ideal torques. We calibrated muscle torque-generating functions for a 3D full body model using dynamometry measurements. An ideal-torque-driven and muscle-torque-driven model were tested in unassisted and walker-assisted gait predictions with different configurations. Results suggest that incorporating muscle torque actuators improve the physiological accuracy of simulations while not reducing significantly computational efficiency. Future work will focus on implementing calibrated lower body muscle models in gait predictive simulations to support clinical decisions.