Nonprehensile Riemannian Motion Predictive Control

Hamid Izadinia Byron Boots Steve Seitz

Top row shows the rgb snapshots of the point cloud stream input to our model which we use to recompose the scene into a simulated 3D scene (middle row) and generate look ahead motion field (bottom row).


Abstract


Nonprehensile manipulation involves long horizon underactuated object interactions and physical contact with different objects that can inherently introduce a high degree of uncertainty. In this work, we introduce a novel Real-to-Sim reward analysis technique, called Riemannian Motion Predictive Control (RMPC), to reliably imagine and predict the outcome of taking possible actions for a real robotic platform. Our proposed RMPC benefits from Riemannian motion policy and second order dynamic model to compute the acceleration command and control the robot at every location on the surface. Our approach creates a 3D object-level recomposed model of the real scene where we can simulate the effect of different trajectories. We produce a closed-loop controller to reactively push objects in a continuous action space. We evaluate the performance of our RMPC approach by conducting experiments on a real robot platform as well as simulation and compare against several baselines. We observe that RMPC is robust in cluttered as well as occluded environments and outperforms the baselines.

Keywords: Riemannian Motion Policy (RMP), Nonprehensile manipulation, Model Predictive Control (MPC), Closed-Loop Control, Real-to-Sim reward analysis, RMPflow, real RC-car robot, fully automatic object-level scene recomposition, point cloud input, YCB dataset.


Real World RC-car Robot


Real world RC-car robot action trajectory generated by proposed RMPC. Each row shows a trial with a different initial location of the target object, other objects as obstacles. The goal location is indicated with a green dot.




Fully automatic object-level scene recomposition


Examples of fully automatic object-level scene recomposition from the point cloud. The first and second rows show the side view and the top view of the captured point cloud. The third row shows our object-level scene recomposition.




Bibtex

            @inproceedings{Izadinia-RMPC-ISER2020,
                title = "{Nonprehensile {R}iemannian Motion Predictive Control}",
                author = "{Izadinia, Hamid and Boots, Byron and Seitz, Steven M}",
                booktitle = "International Symposium on Experimental Robotics ({ISER})",
                year = {2020}
            }