The rest of this dissertation is organized as follows. In the next chapter we review previous work in visual robot control. Chapter 3 contains a tutorial on visual servo control. Its main purpose is to point out how little is required to implement a basic visual control system. This is a point that otherwise might not come across in a thesis size publication. In Chapter 4 we describe the uncalibrated viewing model. Suitable visual appearance and feature measures are discussed in Chapter 4.1 and in Chapter 4.2 we show how to estimate visual motor models based on any of these measures. Chapter 5 describes how to achieve convergent and stable control using the (locally) estimated visual-motor model. Properties of the combined visual-motor model acquisition and control system are evaluated experimentally in Chapter 5.5. Another use for visual-motor models is to visually predict the outcome of a motor action. In Chapter 6 we show how to do that both on-line, augmenting a live video stream while acquiring the model at the same time, and off-line to generate movies based on a previously learned model. Chapter 6.3 shows visual simulation examples with robot and human arms. In the examples with a human arm, the visual-motor models are estimated based on observation only.
The last part of the dissertation, Chapter 7, deals with how to solve whole complex tasks using uncalibrated visual control. Traditionally, task specification and trajectory planning are done in a world coordinate frame, and are mapped to visual and motor frames using a-priori calibrated global transformations. The global world coordinate frame does not exist in the uncalibrated system. We show how task specification and planning instead can be done very naturally using visual alignment specifications in image space, and, only when necessary, using local object-centered Cartesian frames that can be obtained on-line during the manipulation. The usefulness of this approach is demonstrated experimentally in Chapter 7.5, where we show how to make our system solve a child's puzzle and install an ordinary light bulb.