15-494 Cognitive Robotics Final Project:
Cooperative Form Closure and Manipulation of Large Objects
Felix Duvallet, Nicholas Heckman
{fduvalle, nth}@andrew.cmu.edu
The Robotics Institute
Carnegie Mellon University
5000 Forbes Ave
Pittsburgh, Pa 15213

I. Problem Description and Introduction

Manipulating large objects with a single mobile robot is a difficult task. The object is too large to push in a given direction or to rotate accurately. A single mobile base would have to accurately find the center of mass of a large object to push it, or find the end of the object to rotate it. Any error can lead to unintended results. Using multiple mobile bases to enclose and manipulate an object eliminates any need to determine the exact center of mass or endpoints of the object for pushing or rotation. This project demonstrates cooperative form closure and manipulation of large objects using two mobile bases. Cooperative form closure is the ability to secure an object from moving by using the object's geometry and the form of the manipulators. This differs from force closure, where force applied on the object by manipulators and friction are used to secure the object, without utilizing the object geometry. In this project, we use several robots to manipulate a large object by achieving form closure.

Figure 1: Form closure vs. friction closure

Figure 2: Two Sony Aibo ERS-7 robots moving a large object

II. Approach

We use two Sony Aibo ERS-7 robots as a developmental testbed for our solution. The object that we manipulate is a Styrofoam block 19 inches long, 3 inches high, and 5 inches wide. A 4x1.5x5 section block is cut away from the larger Styrofoam block to allow the robots to achieve form closure. The top face of the Styrofoam block had a strip of pink tape running across the long axis, and two pieces of green tape running across the short axis of the top face. These pieces of tape are used for robot localization.

We will create the localization system that will allow the dogs to determine its position and orientation relative to the block. By default the dogs attempt to maintain a constant distance from the end of the block and a constant orientation with respect to the long axis of the block. Distance is determined by looking at the green piece of tape on the block through the robot's camera. The robot computes the length of the tape in camSpace. The longer the perceived length of the tape, the closer the tape is to the robot. Using proportional control, the robot attempts to keep the length of the tape a constant length. Orientation is determined by viewing a pink piece of tape that is placed along the long axis of the block. When viewed in cam space, a pink line that is in alignment with the robot's X-axis runs through the center column of the camera image. As the orientation of the line deviates from the robot's X-axis, the centroid of the line likewise deviate from the center. We use a proportional control to maintain the pink centroid in the center of the image. We have the robots maintain a pose with respect to the block such that the robots can move their end of the block slightly by strafing or pushing.

Figure 3: Top view of the block and localization markers

Figure 4: Two dogs cooperatively grasping and pushing the object


We achieve form closure by localizing the block with respect to the robot. We manipulate the block by having the robots attempt maintain different positions and orientations with the block. As one robot moves the block to achieve its goal orientation, it changes the position and orientation of the block that the second robot perceives. The second robot then moves the block to achieve its goal orientation. By using this approach, all manipulation coordination is implicitly performed. No inter-robot communication occurs.

Figure 5: View from the Aibo's camera. Block is aligned in this image

Figure 6: Aibo's view. Block is off-center

Figure 7: Lines extracted from the camera image

III. Results

These are demonstration videos of closure, pushing, and a more advanced box manipulation.
Approach and grasp closure of a large rectangular block
Push of rectangular block
Push+Rotation of block (move around corner)

code directory

IV. Summary and Future Work

We have shown a method to achieve form closure of a rectangular block using two mobile robots. Localization is performed using simple markers. Additionally, we have demonstrated coordinated manipulation of the block by using push and rotate motion biases, using only implicit robot-robot coordination. In the future, we plan on using global localization to allow for exact positioning of the object.


2007 Felix Duvallet, Nicholas Heckman