Prior research on humans undergoing a forward fall has revealed that the deployment of a rollover strategy along the longitudinal axis can lower the impact force experienced on the hand to effectively reduce wrist injuries. Yet, analogous research for humanoids has received scant attention. To address this research gap, in this work, we consider the optimal design, implementation, and examination of a rollover strategy–similar to the one for humans–for a humanoid robot by employing a differential dynamic programming (DDP) approach.

Previously, researchers have designed a parachute landing fall (PLF) motion heuristically by considering only one side of a humanoid. However, such a model cannot be reliably applied to a full humanoid without considering actual contact environment in trajectory optimization. We consider parachute landing based on a full biped robot with a rigid contact model, utilizing the differential dynamic programming (DDP) method.

To improve on prior works, we consider a bidirectional long short-term memory (BLSTM) network, which makes use of historical measurements of system states as inputs, to effectively predict fall probability in real time. Through extensive simulation experiments, which utilize external forces with random magnitudes, directions, locations, and times of application, we demonstrate that the proposed BLSTM network can robustly predict fall events.