Autonomous Target Recognition and Localization for Manipulator Sampling Tasks

Abstract

Future exploration missions will require autonomous robotic operations to minimize overhead on human operators. Autonomous manipulation in unknown environments requires target identification and tracking from initial discovery through grasp and stow sequences. Even with a supervisor in the loop, automating target identification and localization processes significantly lowers operator workload and data throughput requirements.

This thesis introduces the Autonomous Vision Application for Target Acquisition and Ranging (AVATAR), a software system capable of recognizing appropriate targets and determining their locations for manipulator retrieval tasks. AVATAR utilizes an RGB color filter to segment possible sampling or tracking targets, applies geometric-based matching constraints, and performs stereo triangulation to determine absolute 3-D target position.

Neutral buoyancy and 1-G tests verify AVATAR capabilities over a diverse matrix of targets and visual environments as well as camera and manipulator configurations. AVATAR repeatably and reliably recognizes targets and provides real-time position data sufficiently accurate for autonomous sampling.