An Evaluation of Particle Filters for Contact-SLAM Problems

The contact-SLAM problem is a broad class of grasping and manipulation problem and it is very important to robotic manipulation tasks, particularly when contacts are likely to be intermittent. Several researchers have developed particle filters for C-SLAM problems that estimate the state of manipulated objects, some geometric properties, and their contacts, but the effects of various designing issues are obfuscated by implementation details and are not discussed thoroughly. In this paper, we study the C-SLAM problem restricted to the contact normal direction to understand fundamental modeling choices through analysis of real and simulated experiments. We compare the results of both rigid body and compliant body model with several sources of noise to give users guidance in implementing particle filters for C-SLAM problems.