Alan D. Christiansen

Manipulation planning for empirical backprojections

A method of computing manipulation plans is introduced. The empirical backprojection method uses observed state transitions to derive predictions of action effects and links these predictions to form manipulation plans. The data-intensive approach is in contrast to manipulation planning approaches that uses analysis of task mechanics to generate predictions. The empirical backprojection method is described, and three example tasks are used to illustrate the method. The assumptions underlying the method are then compared to the assumptions required by previously published analytical techniques.<<ETX>>

Experiments in robot learning

Publisher Summary This chapter reviews two robots that can learn how the world behaves in, and improves their own performance over time based on the information gathered. The underlying architecture of such robots would be of the abstract agent variety and implemented in CommonLisp. The abstract agent defines a framework for defining and testing different learning agents. Also, the architecture of the abstract agent allows a new agent to be created by plugging in a new module or recombining old modules. The experimental task domain defined for the robots is the operation of a tilting tray. A robotic manipulator holds a square tray, bounded by four vertical walls, which may be tilted at any angle desired. An object, typically a square block, slides freely in the tray. By a judicious choice of tilting sequence, the robot is able to position the object in the tray. The chief complication of the tilting-tray occurs when the object hits a wall. Now, one of the experimental robots is proposed to have been fitted with the rote agent performs rote-learning of tilt motions mechanics while the other with the inductive agent, which does a simple version-space learning to generalize in the space of tilt-angles. Both agents use exhaustive search to construct plans. When no such plan can be constructed, a single random tilt is chosen.

Probabilistic analysis of manipulation tasks: a research agenda

The problem of manipulation planning in the presence of uncertainty is addressed. The worst-case planning techniques introduced in Lozano-Perez, Mason, and Taylor (1984) are reviewed. It is shown that these methods are limited by an information gap inherent to worst-case analysis techniques. As the task uncertainty increases, these methods fail to produce useful information even though a high-quality plan may exist. To fill this gap, the probabilistic backprojection, which describes the likelihood that a given action will achieve the task goal from a given initial state is presented. A constructive definition of the probabilistic backprojection and related probabilistic models of manipulation task mechanics is provided. It is shown how these models unify several past results in manipulation planning. These models capture the fundamental nature of the task behavior, but appear to be very complex. Methods for computing these models are sketched. Efficient computational methods remain unknown.<<ETX>>