Bit-depth sensitivity.mp4

The video compares closed-loop rollouts under 4-bit (top row), 6-bit (middle row), and 8-bit (bottom row) quantized state-pattern inputs over a 10 s horizon from the same initial condition. Each row shows three synchronized views: (left) the encoded 2D input pattern displayed to the optical actor, (middle) the rendered pendulum motion in the simulator, and (right) the corresponding CCD-captured optical readout. As bit-depth decreases, quantization not only reduces grayscale resolution but also induces larger visible changes in the measured optical readout structure, which correlates with degraded stabilization performance (4-bit fails to stabilize, 6-bit shows intermittent loss of balance, and 8-bit achieves sustained balance).