Monte-Carlo Testing for AUV Planning Software
Zeyn Saigol · 2008
Constraint-based planning systems, especially those used for real-world applications, have a very large space of possible states. This makes them very hard to test, as only one path through the state space can be tested at once, and it is tricky to manually select a small set of paths that will thoroughly test a system. This paper describes my implementation of a Monte-Carlo testing system for the T-REX planner, which was developed at MBARI for AUV control. The test harness performs repeated runs of the planner, forcing each run through a series of randomly sampled states, in an attempt to uncover bugs in the system that manual methods may fail to detect.