Even the most robust character or robot is bound to fail under some extreme circumstances. We would like to have some fall management mechanism to deal with such situations. In this work we develop a basic falling controller for a humanoid character. The controller used a low level quadratic programming based torque controller which allows us to specify simple goals such as a target center of mass velocity or target orientation. We developed several fall damage measures such as impact force, location of impacts, and final fall location. We use these error measures to optimize the parameters of our basic falling controller ensuring we can control falls to cause the minimum amount of damage.
Ukemi-like strategy with no arms
Fall without our reflex phase
Simulation and Control of Humanoid Rolling and Falling Behaviors
David F. Brown, Master's Thesis, 2012