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

Robust Perception and Control for Humanoid Robots in Unstructured Environments Using Vision

Geoffrey Taylor, PhD Thesis, Monash University

View Manuscript (PDF, 9.5 MB)


Humanoid robots promise to fulfill the role of universal domestic aid, with particular benefit for the care of the elderly and disabled. This thesis develops the fundamental visuomotor skills required by a humanoid robot to perform useful manipulation tasks. The working hypothesis is that the requirements of autonomous operation in an unstructured, unpredictable environment, based on supervisory commands involving unknown objects, and with robustness to calibration errors can be met using visual sensing. For each new task, the robot first acquires a dense 3D image of the scene using a novel stereoscopic light stripe scanner. The range map is segmented into geometric primitives, and simple objects (cups, bowls, boxes, balls, etc) are autonomously identified and modelled. To manipulate objects, the arms of the robot are controlled using visual feedback in a framework known as position-based visual servoing. During grasping, the objects are reliably tracked through clutter and occlusions by exploiting multimodal cues (colour, texture and edges).

The pages on this site provide video clips, 3D models and other material to supplement the experimental results for light stripe scanning, object modelling, 3D tracking and simple manipulation tasks presented in the thesis. The multimedia extensions are arranged according to relevant chapter, and can be accessed using the menu on the left or the links below:

The following additional extensions are not covered in the body of the thesis:

Video clips are provided in MPEG-1 format, and light stripe scanner data and object models are provided in VRML 97 (Virtual Reality Markup Language) format, which requires a suitable viewer (see list below). In addition, each VRML file is accompanied by an MPEG video "fly-through" to visualize the 3D model without a VRML browser.

Free VRML viewers:


I am currently working as a Research Fellow at the ARC Centre for Perceptive and Intelligent Machines in Complex Environments (Feb - Aug, 2004). My contact details during this time are listed below. Click here for an electronic version of my current Curriculum Vitae (PDF, 62 Kb).