Fall recovery after falling over is a critical feature for biped robots that work in the real world. We propose a method by considering robot joint angle trajectory as a multivariable function of time and orientation of robot root links. Compared with the conventional method, our method required less time and human effort, as the joint trajectory was autonomously optimized via evolutionary search in a dynamic simulator. In addition, the gap between simulation and real-world experiments could be narrowed by adding push disturbances when performing evolutionary search in the dynamic simulator. Using the proposed method, we successfully realized the fall-recovery feature on life-sized biped robot that was equipped with shell protector CHIDORI-F as depicted in Fig. 1. Combining the aforementioned fall-recovery feature and previously developed fall-tolerant robot hardware, fall detector and fall-damage-reduction motion, we realized a biped robot platform that could tolerate falling, recover from all the fallen states, and continue working for considerable time without a tether.