Autonomous Gas Leak Detection Robot — ESP32 + Live

Power: Two 2S LiPo packs are clunky. A single 3S (11.1V) would feed the L298N directly — simpler wiring, more capacity.

Gas sensors: MQ-135 covers CO2, NH3, benzene, smoke, and alcohol in one package. Replacing both MQ9s simplifies wiring and broadens coverage.

Mapping: Current dashboard map is dead reckoning from MPU6050 yaw — drifts badly over time. Real SLAM would make it useful for actually locating a leak source.

Logging: Nothing is persisted right now. SD card logging would store timestamped readings even when WiFi is out of range.

Why two batteries: One 2S LiPo powers motors through L298N. A second powers all logic through the PSU module. Running everything from one battery caused ADC spikes every time a motor started — motor current dragged supply voltage down, showing up as false gas readings. Separate supplies fixed this completely.

Why GPIO15 as a power output: GPIO15 is set HIGH in software at boot and stays there — becomes a 3.3V source pin for sensors. It's a strapping pin that must be HIGH at boot anyway, so it's a natural choice.

Why input-only pins for sensors: GPIO34, 35, 36 (VP) have no internal pull resistors and can't be outputs. This makes them ideal for analog inputs — no accidental output state, no pull affecting ADC readings.

This entire project was designed, wired, coded, and tested in a single 24-hour hackathon session.

The biggest technical fight was the MQ9 sensor supply voltages. Both sensors read garbage for the first 3 hours. The CO sensor was on 5V — outputting saturated readings regardless of actual gas. Fix: CO sensor needs 3.3V, LPG sensor needs 5V. Once sorted, everything made sense.

The ESP-CAM integration was surprisingly smooth. Instead of a serial link, we made it join the ESP32's own WiFi AP and stream over HTTP. The dashboard embeds the stream URL as an image source. Simple and it works.

IR wall-following ran 8+ minutes without collision in the test corridor.

Code: https://github.com/Nikhil-A-E/Gas-Leak-Detection-Bot
Video: