Dr. Nathalie Cabrol

December 19, 2006

"Science on the Fly: Enabling Science Autonomy During Robotic Traverse with Application to the Study of Life in the Atacama Desert of Chile"

CMU110079-150783

Currently, rovers can travel tens of meters per sol, but the next generation of rovers is predicted to be able to traverse ten or more times as far. This capability presents a situation in which the rover will be able to travel to places about which it has no information and which it was not able to see at the beginning of the sol. This presents us with not only an exciting opportunity but also several challenges. This situation will require the rover to use methods for performing effective science when the rover has left the area that was initially visible and is out of contact with scientists.

The "Science on the Fly" software consists of two principal parts, the science observer and the science planner. The science observer acts as the rover's eyes and ears. The science observer interprets sensor data to find possible targets of scientific value. The other part, the science planner, takes the observations made by the science observer and plans experiments that would be of maximum scientific value.

Science Observer

The Science Observer serves two major roles: identifying targets of interest and categorizing them into useful groups. For example, the most prominent targets for geological study are rocks. When the science observer examines an image, it looks for rocks using a "machine learning" algorithm that has been trained ahead of time to recognize rock-like features. Then, it autonomously categorizes those rocks into groups based on data it has seen before. In this manner, the Science Observer can recognize novel rocks as well as perform automatic geological analyses. Besides rocks, targets of scientific interest could include lichens or patches of soil.

Science Planner

The Science Planner enables the rover to react to new science opportunities as it moves into unexplored areas. The Science Planner accepts priorities from the science team, such as "carbonate rocks are important to sample." Then, when the Science Observer detects a high-priority, interesting feature, the Science Planner tries to make a plan for getting more useful information about the feature. For instance, it might move closer and examine the feature with its fluorescence sensor to look for signs of life. It is not always easy to generate this kind of plan because examining every interesting feature would take far more time and energy than the rover has available.