Robust Perception and Control for Humanoid Robots in Unstructured Environments Using Vision
Geoff Taylor, PhD Thesis, Monash University






Chapter 7
Integration

In chapter 7, perception and control are integrated to realize a domestic humanoid robot capable of performing interactive tasks with unknown objects in an unstructured environment. Two domestic tasks are implemented to evaluate the performance of the robot, and supplimental results are presented below.
 

Experiment 1: Grasping an Unknown Object (Section 7.3)

In this first experiment, Metalman is given the task: "fetch the yellow box", ie. retrieve an object from a cluttered scene that has only been specified by generic class (box) and feature (yellow). No other information is initially known (apart from prior autonomous calibration of the light stripe scanner and approximate hand-eye transformation), and the manipulator is initially outside the field of view. The VRML and MPEG results presented below show the initial raw colour/range scan, extracted objects, and grasp execution as seen through the right camera.
 

Raw Colour/Range Scan (Figure 7.7(a))
Extracted Objects (Figure 7.7(d))
Object Tracking and Grasp Execution (Figure 7.9)

Experiment 2: Pouring Task (Section 7.4)

In the second experiment, Metalman is required to grasp a cup (filled with rice) that has been interactively selected by the user and pour the contents into a bowl, under similar conditions to those described above. This experiment simulates the type of interaction that might be necessary when ambiguous or incomplete task specifications are encountered, ie. Metalman has insufficient information to determine which cup contains the rice. As before, the VRML and MPEG results provided below show the initial raw colour/range scan, extracted objects, and task execution as seen through the right camera.
 

Raw Colour/Range Scan (Figure 7.11(a))
Extracted Objects (Figure 7.11(d))
Interaction, Object Tracking and Grasp Execution (Figure 7.13)