Contribution: Kuan-Ting Yu and Alberto Rodriguez

Realtime Tactile-aware Object Pose Estimation in Planar Pushing using iSAM

Kuan-Ting Yu1 and Alberto Rodriguez2
1Computer Science and Artificial Intelligence Laboratory — Massachusetts Institute of Technology
2Mechanical Engineering Department — Massachusetts Institute of Technology

Planar pushing is an important object manipulation strategy. However, pushing process has nonnegligible uncertainty, so reactiveness is needed to complete an unstable pushing task. To do so, estimating the current pose of the object in real time becomes important. In this paper, we are inspired by a popular framework for solving robot localization problem – iSAM (incremental smoothing and mapping) and adapt it to provide a fast and flexible solution. We tailored its measurement cost, and motion cost for the pushing scenario. On the measurement, we fuse information from cameras and that from contact sensors. The cameras provide global pose information but are noisy in general, whereas contact sensing is local but the measurement will be more accurate relative to end effectors. Therefore by combining the two we can exploit the advantage from both sides. We use a well-instrumented setup to evaluate the algorithm with different object shapes
and on different surface materials to show it’s applicability.