Vision-based homing of an AUV

ENDURANCE performed scientific missions totaling more than 10 weeks in West Lake Bonney in the Antarctic Dry Valleys. Lake Bonney is perennially covered with ice about 3 meters thick. During a typical mission, the vehicle was deployed into the lake through a melt hole in the lake ice, traveled a distance of several hundred meters to sample the biogeochemical properties of lake water, arrived at the melt hole, and rose to the top for downloading data and recharging. Logistic difficulties of melting a 1.5 m. diameter hole in ice 3 m. thick in extreme cold weather meant that we could melt a single hole only slightly larger in diameter than the robot. The robot had to deployed and recovered through the hole.

ENDURANCE in West Lake Bonney. (a) Aerial view of Lake Bonney (b) ENDURANCE being flown to the field-site by a helicopter as a sling load (c) Melting a hole in the lake ice for deploying the vehicle into the lake (d) ENDURANCE being deployed through the melt hole. A blinking light source was suspended about 1.5 m. above the melt-hole.


The vehicle's dead-reckoning based navigation system was good enough to get it to within 5 meters of the melt hole after navigating a distance of several hundred meters. We developed a visual homing algorithm that could detect and identify a blinking light source, and then use this light source to locate and ascend up a melt-hole. A blinking light source was chosen after experimentation to avoid aliasing with ambient light. The vehicle first initiates a search for the light source. Once it detects a blinking source, it locks on to it and ascends up the melt hole. The pictures below show the algorithm in action.

Spiral search and ascent. Once the vehicle is at its nominal home position, it starts moving in a spiral pattern while maintaining its depth. Once it detects a blinking light source, it rises up while keeping itself centered on the light source.

Images taken by upward-facing camera during visual homing in West Lake Bonney. (a) The vehicle approaches the melt-hole. The under-surface of ice is seen. (b), (c) The light-detection algorithm identifies a potential target as the vehicle passed beneath the melt-hole. (d),(e),(f) The blinking light is identified as the target and the vehicle ascends up the melt-hole.