Exploring the Contact Space to Plan Robot Motions under Geometry Uncertainty Constraints

The work presented on this paper has been derived from previous work around the area of ne motion assembly planning. However, the method that we present can be applied to solve diierent robot motion planning problems. The robustness of the trajectories produced by a robot motion planner depends on both, the models used to represent the uncertainty, and the strategies used to deal with it. A robot motion planner receives a geometric description of the manipulated object and its environment. This description is called the \nominal world" and is just an approximation of the mechanical world where the robot operates. The success of a robot manipulation program strongly relies on the information that the planner extracts from the nominal world to reduce the uncertainty, and how this information is combined with sensing operations in a control program. This paper presents a local approach to plan robot motions by exploring the contact space. A potential eld function deened over both, the free space and the contact space is applied to generate the motions. Potential eld methods are normally applied to compute collision-free trajectories by attracting the robot to the goal connguration and pushing it away from the obstacles. The particularity of this approach is that the potential eld method is used to explore the contact space. The contacts against the obstacles are used to control the motions of the robot reducing its position uncertainty. The main drawback of the potential eld methods is that they can lead the search to local minimums. A search function is used to determine sub-goal conngurations within the non-explored contact space to correct them.