Investigating the effect of phase sensitivity and cost function selection in predictive simulation of human stumble...

Falls initiated by stumbling are a major cause of injury, yet experimental measurement is limited by ecological validity and safety concerns. While predictive neuromechanical simulations provide a powerful and safe alternative for studying stumble recovery, its utility relies entirely on identifying the underlying control objectives required to accurately replicate human-like recovery mechanics. To address this, we systematically examined recovery behavior using a predictive neuromechanical simulation framework with a sagittal-plane musculoskeletal model. The swing foot of the model was obstructed at 15 distinct gait phases (10-90% swing) and recovery was optimized under four physiologically-based cost functions: cost of transport (CoT), muscle activation (MA), head stability (HS) and ground reaction force impact (GRFI). Recovery strategy was strongly phase-dependent: early-swing stumbles predominantly resulted in elevating strategies, while late-swing stumbles resulted in lowering strategies. Mid-swing was identified as an inherently unstable region where recovery consistently failed. Minimizing GRFI best replicated the elevating strategy (minimal trunk flexion) in early swing, while maximizing HS best replicated the lowering strategy in late swing. Crucially, minimizing CoT or MA alone failed to predict successful, human-like recoveries. Stumble recovery is governed by phase-specific trade-offs where safety (impact control and head stabilization), not energy economy, dominates control priorities. These findings advance the neuromechanical understanding of reactive balance and inform fall-prevention strategies.