While building my vision-based following robot, I encountered a quite funny phenomenon with its headlights. The robot uses led strips to highlight its path whe

While building my vision-based following robot, I encountered a quite funny phenomenon with its headlights. The robot uses led strips to highlight its path when it detects that the luminosity from the camera is not enough using simple PID control loop. In normal situation, it is working as intended making the robot more robust to illuminations change. But every time the robot crashes into an object or wall, it flashes its headlights like if it is trying to call for attention. What is happening is when it crashes the objects are usually really close to both the leds and the camera so suddenly it hide exterior light and it's very dark for the camera so the robot turns the leds on, but because the objects are so close a lot of light is reflected back and suddenly it's too bright so it turns the leds off and the cycle repeat : My "auto-exposure" control loop couldn't keep up and is no longer stable. The robot had inadvertently acquired a very crude active-light obstacle avoidance / distance sensing system. To add to the complexity of modern headlights, you can now envision that to efficiently exploit this sensor you can plug in a neural network to actively control the headlights to give itself a sense of depth perception. Add multiple robots into the mix and let them evolve a headlight communication strategy/protocol... Here was a little fun dose of complexity to tingle your light-bulb...