This report presents the development and testing of a control framework to enable Bolt, a bipedal robot developed by the Open Dynamic Robot Initiative (ODRI), to walk on flat terrain. The initial approach utilized a control pipeline combining Divergent Component of Motion (DCM) Model Predictive Control (MPC) with Whole-Body (WB) MPC, coupled with a Riccati-based low-level controller. Despite promising simulations in PyBullet, sim-to-real transfer issues prevented stable control on the physical robot. This work, focused on the MPC-based approach and simulations in PyBullet, was accepted for publication at the IEEE International Conference on Humanoid Robotics 2024. To address these limitations, we implemented Reinforcement Learning (RL) techniques using the Constraints as Terminations (CaT) project framework, originally developed for quadrupedal robots. Through RL, the Bolt robot achieved walking behavior in real-world tests, which had not been possible with MPC alone. However, the resulting motion was unstable and difficult to control, highlighting a persistent sim-to-real gap. In investigating this discrepancy, we developed a tool to examine and refine the robot’s inertia tensors, identifying potential inaccuracies in the physical model that could affect real-world control. This analysis underscored the challenges in achieving stable bipedal locomotion and suggests that precise model calibration is critical for robust control deployment.