Wearable robots are becoming increasingly common, in both research laboratories and the industry, due to their significant potential benefits in rehabilitation engineering, assistive robotics, ergonomics, and power augmentation. Thus far, design and control of these devices have primarily relied on exhaustive experimental procedures. Alternatively, combined predictive simulations of device and human musculoskeletal mechanics offer a promising approach to decreasing necessary human subject experiment scenarios and cost. In simulation, the device parameter space can be explored to determine the most promising design solutions and parameter values, which, in turn, can inform the human subject experiment design. This dissertation focuses on building a framework for combined musculoskeletal and exoskeleton dynamics for walking. In the framework, the actuation profiles of body muscles are optimized using a single-shooting method. The single-shooting method facilitates convenient consideration of human musculoskeletal system models with varying levels of complexity, various exoskeletons and controllers, and different objective functions. High-throughput computing resources are employed for the computationally-intensive optimizations in this framework. The proposed framework is used for study and design of passive exoskeletons for reducing the metabolic energy expenditure during walking. The simulation results suggest that elastic elements acting in parallel with lower–limb uniarticular muscles can reduce the metabolic cost of walking by up to 28%. These results support the use of predictive simulations as a tool for the study and conceptual design of exoskeletons and can accelerate device and control development.