Bridging Discrete and Continuous Interfaces to Generate Adaptive Gait Synthesis for Humanoid Robots

As humanoid robots become increasingly prevalent globally, from standardized platforms to cutting-edge work by groups like Boston Dynamics, the need for adaptable and intuitive bipedal locomotion control has grown. This work revisits the Darwin OP, a legacy humanoid platform, to propose a modular real-time walking controller capable of enabling dynamic, omni-directional locomotion through both binary (keyboard) and analog (joystick) inputs. The controller features smooth acceleration, directional blending, and momentum-preserving transitions that support lifelike and responsive gait behavior. Unlike rigid state machines or pre-trained reinforcement learning models, our system interprets user input to modulate stability and motion naturally and flexibly. We demonstrate that even older robotic platforms can achieve modern standards of motion responsiveness and adaptability through thoughtful control architecture, highlighting a pathway for repurposing legacy systems in contemporary robotics research.