To effectively capture the performance of an advanced local planner for a sailboat in a way that can be presented in a conference paper, a single comprehensive test is required. This post contains a working definition of the test that I will use in my conference paper.
It’s worth noting that in an earlier post, behaviour specific scenarios, I defined a large set of test scenarios for evaluating sailboat behaviour in a localized setting. Though valuable, that testing suite does not concisely capture the performance of a planning solution and instead is more valuable for algorithm development.
Overview
The target state space of my current research explicitly defined in this post and continues to be used in this test.
In order to comprehensively capture the performance and validity of a fully autonomous sailboat controller we will take from the methods that have been developed for evaluating humans completing the same task.
Course
Explicitly, the test will mimic the Triangle-Windward-Leeward Course as defined by World Sailing in the Racing Rules of Sailing 2017 - 2020. Since mark-roundings are out of the scope of the challenge being evaluated, involve reaching waypoints arranged in the shape. The controller must simply get the boat within of each waypoint before heading to the next. The initial pose will have the boat going close hauled on a starboard tack with a velocity of 0 at the middle of the start line. The last point will be at the middle of the finish line. The distance between points 1 & 3 is 5 km.
Obstacles
In order to test the obstacle avoidance capabilities of the controller, randomly emitted obstacles will be generated from one of the edges of the “world” at a period of 10s.
Last updated: March 7th, 2017