Collision-Based Contact Mode Estimation for Dynamic Rigid Body Capture

Abstract

This paper proposes real-time collision-based contact mode estimation with only a force-torque sensor for capturing a moving rigid body. The contact modes are defined for determining when to generate the signal to close the robotic hand for establishing object closure. In our particle filter approach, collision-triggered filter is used to determine the contact mode with the least amount of computation. Brach's collision model is used for our collision model-based approach for a rigid body because it is computationally light-weighted and enables the sampling of three collision properties for the particle filter. The validity of our method is experimentally demonstrated by achieving the highest success rate using the reasonable computation resources required (average of 3.9 milliseconds and worst of 6.1 milliseconds with our setup), and verifying each computation resource (or number of particles) based on the size of motion estimation error in the pre-capture phase.