Evaluation of an interactive technique for creating site models from range data

Abstract

Recognizing and locating objects is crucial to robotic operations in unstructured environments. To satisfy this need, we have developed an interactive system for creating object models from range data based on simulated a annealing and supervisory control This interactive modeling system maximizes the advantages of both manual and autonomous methods while minimizing their weaknesses. Therefore, it should outperform purely autonomous or manual techniques. We have designed and executed experiments for the purpose of evaluating the performance of our application as compared to an identical but purely manually driven application. These experiments confirmed the following hypotheses: (1) Interactive modeling should outperform purely manual modeling in total task time and fitting accuracy. (2) Operator effort decreases significantly when utilizing interactive modeling. (3) User expertise does not have a significant effect on interactive modeling task time. (4) Minimal human interaction will increase performance on {open_quotes}easy{close_quotes} scenes. Using 14 subjects and 8 synthetically generated scenes, we recorded the task times and pose data and, from them, used analysis of variance (ANOVA) to test a set of hypotheses.