Quantifying balance: computational and learning frameworks for the characterization of balance in bipedal systems

In clinical practice and general healthcare settings, the lack of reliable and objective balance and stability assessment metrics hinders the tracking of patient performance progression during rehabilitation; the assessment of bipedal balance plays a crucial role in understanding stability and falls in humans and other bipeds, while providing clinicians important information regarding rehabilitation outcomes. Bipedal balance has often been examined through kinematic or kinetic quantities, such as the Zero Moment Point and Center of Pressure; however, analyzing balance specifically through the body\u27s Center of Mass (COM) state offers a holistic and easily comprehensible view of balance and stability. Building upon existing boundary-based stability criteria, a balance region (BR) can be constructed in the COM state space (COM position and velocity) by identifying the border of the COM state space within which the system can regain its balance by returning to the equilibrium state (static upright posture). In contrast to many other approaches, the BR considers factors such as subject- or patient-specific actuation limits, perturbation and contact responses, boundary constraints, and other essential components that hold significance in rehabilitation. In this work, a recently developed COM BR method is first extended within an optimization-based computational framework and used to quantify balance with evaluation of its reachable and viable margins in depth. By demonstrating the potential of this method in quantifying balance in healthcare environments, a tele-health protocol is introduced for the remote assessment and rehabilitation of patients, through which balance exercises can be studied within the BR framework. Extracted information can then be relayed to the patient as exercise goals, enabling effective monitoring and guidance for rehabilitation. Next, to incorporate the contribution of neuromuscular factors into the BR, a reinforcement learning (RL)-based framework is used for the development of a real-time, muscle-based balance controller. This controller activates individual muscles within a musculoskeletal model in response to its state, bringing it to equilibrium; consequently, it can be utilized to generate the BR through numerous simulations with varying initial COM states. In addition, altering muscle properties affected by neuromuscular disorders or aging (e.g., muscle weakness, hemiplegia) significantly reduces the size of their respective BRs, providing insights into how balance is affected by physiological changes to muscle and offering a pathway to further study other neuromuscular conditions. Lastly, an experimental study of various balance exercises is performed to demonstrate the feasibility of empirically generating BRs and assessing balance, as well as determining the contributions of major balancing muscles during these exercises through the analysis of muscle activation patterns. The results show that humans are less likely to reach their theoretical BR limits, when an ankle strategy is encouraged, which is in line with the findings obtained from numerically generated BRs. In summary, this work presents compelling evidence that the proposed frameworks for BR generation and analysis can be effectively employed for the quantification and subsequent analysis of balance. This has significant implications for patient diagnostics, monitoring, and rehabilitation, offering the potential for improved outcomes in these areas