Grasping Unknown Objects with a Humanoid Robot

This paper combines the authors’ previous work on a self-calibrated, position based visual servoing framework for a humanoid robot, with a robust laser stripe scanner that can capture registered colour/range measurements of arbitrary objects in ambient indoor light. Using stereo measurements for validation, the laser scanner is robust against sensor noise, spurious reflections and cross talk from other robots. Range data is processed to identify objects of interest in the workspace, which are modelled using simple geometric primitives. The resulting textured 3D models can be used for recognition, tracking and grasp planning. Finally, we present a simple grasp planner that guides visual servoing in the task of grasping a modelled object. The fusion of these components allows the humanoid robot to locate and grasp a class of a priori unknown objects in its workspace. The effectiveness of these techniques are demonstrated on an experimental humanoid platform.