Objective Enable estimation of knee adduction moment (KAM) and knee flexion moment (KFM) under different gait testing conditions via an inertial sensor network. Methods Twelve healthy young male subjects wore eight inertial sensors and walked under different testing conditions (changing foot progression angle, trunk sway angle, step width, and walking speed). A recurrent neural network (RNN) was developed to estimate KAM and KFM with biomechanical features extracted from an inertial sensor network. Results Overall KAM estimation accuracy was rRMSE=8.54% and r=0.84. Overall KFM estimation accuracy was rRMSE=6.40% and r=0.94. Conclusions The model enabled knee joint estimation out of the lab and could potentially serve for gait training and knee surgery outcome assessment.