This study investigates the similarities and differences in the analysis of human walking motion between the traditional inverse dynamics method and the forward dynamics method that employs an Artificial Neural Network (ANN)-based controller. Nine healthy male subjects walked at their preferred speeds while motion capture and ground reaction force data were collected. Inverse kinematics and dynamics analyses were conducted using OpenSim. The ANN-based gait controller was trained via deep reinforcement learning using a two-stage curriculum in forward dynamics simulations. It was first trained for kinematic tracking and then further optimized to minimize torque, power, torque difference, and ground reaction force fluctuations.

The ANN-based controller reproduced joint kinematics with a root-mean-square (RMS) difference of less than 2.7° compared to inverse kinematics in OpenSim. The controller preserved accurate gait kinematics despite reducing joint torques and power. Joint torque profiles showed RMS differences of 0.20–0.23 Nm/kg, comparable to results obtained through optimization-based residual force minimization. Joint power analysis revealed that inverse dynamics in OpenSim underestimated total energy consumption by 0.74 W/kg compared to forward dynamics. This discrepancy was primarily due to residual forces and torques, which accounted for 19.9% of total mechanical power. When residuals were included, the difference in total power between the two methods was reduced to 4.1%. These findings indicate that ANN-based forward dynamics modeling can accurately reproduce human gait while allowing mechanical energy estimation without residual forces. The controller’s adaptability allows for analyzing gait variations under different conditions, with potential applications in rehabilitation and assistive robotics.