The inverted pendulum is a classic engineering problem used to study inherently unstable systems, such as self-balancing robots. We previously developed a low-cost version that successfully balanced the pendulum upright, but it suffered from timing jitter caused by MicroPython programming and significant quantization noise that limited the control speed. This project improved the system to make the control faster and smoother. We eliminated the timing jitter by transitioning to a real-time C environment that runs faster and with consistent timing. To reduce quantization noise, we replaced a simple backward difference velocity estimate with an adaptive windowing method that dynamically adjusts how much data it uses based on how fast the system moves. Adaptive windowing effectively smoothed quantization noise without slowing the system’s reaction speed. We validated these upgrades using a custom program that automatically moves the system and logs real-time balancing data. These improvements increased the stable control frequency to 2 kHz and resulted in audibly smoother motor operation with reduced current spikes. The improved design is an open-source, affordable platform for teaching and research that enables further investigation in control system engineering and machine learning. We plan to share the design as an alternative to expensive commercial equipment.
