Stability Analysis of Aged Locomotion in Physically Simulated Virtual Environments

Human locomotion simulations have broad applications in virtual reality systems, and using deep reinforcement learning the intelligence of character control policies can be further enhanced. However, there are few physics simulations or simulated applications that are specifically designed for older adults as a demographic group. In this work, we consolidated a diverse set of aged locomotion data, and successfully trained control policies through deep reinforcement learning for virtual characters to navigate in simulated environments imitating aged gaits. We further designed and conducted a series of “push”, “trip” and “slip” fall-inducing experiments on the older virtual characters to analyze their locomotion stability without jeopar-dizing real human subjects. We report the initial results from the experiments which could further provide safety insights to prevent falls for older adults navigating similar environments in the real world.