A UAV system for inspection of industrial facilities

Develops a small UAV for inspecting enclosed industrial spaces and flies it autonomously inside a power plant boiler. With no GPS, a lightweight visual-inertial navigation system using only an onboard MEMS IMU and a stereo camera provides state estimates, and model-predictive control allows accurate flight close to the boiler wall. It also presents an integrated, tightly calibrated visual-inertial sensor whose motion estimates drifted only a few centimetres over a flight against laser-tracker ground truth.

Key findings

• Autonomous flight demonstrated inside a power plant boiler
• State estimation used only an onboard MEMS IMU and stereo camera pair
• Integrated visual-inertial sensor drifted only a few centimetres over a flight versus laser-tracker ground truth

Limitations

• Abstract only reviewed
• Early prototype results; the integrated sensor was preliminary
• Dust and darkness handling not quantified in the abstract