Human Digital Twin for Realistic Fall Simulation Using the Behavioral Foundation Model Meta Motivo

Abstract Simulating human movement is essential for biomechanics, ergonomics, and machine learning applications. However, certain motions, such as falls, sudden collapses, or hazardous workplace incidents, are difficult to replicate in real conditions. Meta Motivo, a behavioral foundation model, provides a simulation framework based on reinforcement learning to simulate realistic human movements in a physics-based environment. It can be prompted to solve unseen tasks such as motion tracking, pose reaching, and reward optimization, while generating realistic simulations expressing human-like behavior. This work explores its potential in simulating challenging scenarios through a case study on falls, a critical issue in elderly care and workplace safety. Several strategies for simulating falls are presented, including custom reward functions, reaction time adjustments, and environmental modifications (e.g., slippery surfaces, obstacles, and external forces). This paper also presents two plugins developed to exploit these simulations: one for biomechanical analysis and another for the generation of a dataset for machine learning pose classification models. Although the results are promising, future refinements could further enhance the realism of falls and close-to-ground movements. However, this study highlights the broader applicability of Meta Motivo in replicating movements that are otherwise challenging to replicate experimentally.