Unified comparison of spinal locomotion control architectures in neuromechanical simulations

Neuromechanical simulations provide a powerful framework for investigating how neural control architectures generate and regulate human locomotion. Numerous biologically inspired locomotion controllers have been proposed, including reflex-based, central pattern generator (CPG)-based, and muscle synergy-based models. However, direct comparison across studies remains difficult because of differences in musculoskeletal models, optimization methods, and evaluation protocols. Here, we implemented four representative locomotion control architectures, reflex-based, CPG-reflex-based, muscle synergy-based, and CPG-reflex-synergy-based controllers, within a unified neuromechanical simulation framework to enable controlled comparisons under shared biomechanical and computational conditions. Performance was assessed in terms of (1) agreement with experimentally observed gait characteristics, including kinematics, kinetics, muscle activations, and biomechanical trends across speeds and slopes, and (2) locomotor versatility across speed-slope conditions. The reflex-based and CPG-reflex-synergy-based controllers most closely reproduced experimentally observed gait characteristics, while the CPG-reflex-synergy controller achieved the broadest range of stable walking behaviors across speeds and slopes, followed closely by the reflex-based controller. These findings should be interpreted as comparisons of specific model implementations rather than definitive evaluations of the underlying biological hypotheses. Moreover, because the investigated controllers primarily focused on spinal-level mechanisms for nominal steady-state locomotion, the limited versatility observed in some of the models across broader speed and slope conditions suggests the importance of integrating spinal locomotor mechanisms with supraspinal modulation when modeling locomotion beyond nominal steady gait. To facilitate further investigation, we publicly share the simulation framework and controller implementations.

Existing neuromechanical locomotion controllers have been difficult to compare directly because of differences in simulation frameworks.We implemented four representative spinal locomotion control models (reflex-based, central pattern generator (CPG)-reflex-based, muscle synergy-based, and CPG-reflex-synergy-based) within a unified simulation framework and compared their human-likeness and versatility.The reflex-based and CPG-reflex-synergy-based controllers best reproduced human-like gait characteristics, while the CPG-reflex-synergy-based controller demonstrated the greatest locomotor versatility across speed-slope conditions, followed closely by the reflex-based controller.Because the investigated controllers primarily modeled spinal-level mechanisms associated with steady-state locomotion, their reduced adaptability across broader speed and slope conditions highlights the importance of incorporating supraspinal modulation when modeling locomotion beyond nominal gait.We publicly share the simulation framework and controller implementations to support further investigation of human locomotion control.