Introduction
In robotics, manipulating an object typically relies on precise sensing or a highly controlled environment. However, sliding is inherently unpredictable because the motion of a pushed object depends on its "support distribution"—the exact points of contact between the object and the surface. Since these distributions are often unknown due to microscopic surface irregularities, predicting exact rotation and translation is difficult. This seminal 1986 paper introduces a mathematical framework to provide quantitative bounds on an object's motion, enabling "sensorless" manipulation strategies.
The Bounding Disk and COR Locus
To address the uncertainty of support distributions, the authors model the object as a disk. Any support distribution of a real object can be treated as a valid distribution for a disk that encloses it, allowing the disk's behavior to serve as a "worst-case" bound for the real object.
The core of the method involves finding the Locus of Centers of Rotation (COR). By assuming quasi-static motion (where friction dominates inertia), the authors can identify regions where the COR must lie. These boundaries determine the minimum and maximum possible rotation rates for any given push. By integrating these differential equations, they can find the "worst-case" outcome of a push over a gross distance.
Key Insights: The Tip Line
A significant discovery in the paper is the "Tip Line." The authors found that as the angle of a push ($\alpha$) varies, the furthest possible point of the COR locus traces a straight line perpendicular to the contact vector.
This geometric rule simplifies the calculation of motion bounds. Specifically, the paper provides a concrete formula for the maximum distance of the COR from the center of mass: $r_{tip} = a^2 / (c \cdot \cos \alpha)$, where $a$ is the disk radius and $c$ is the contact distance. This "aha moment" allows engineers to guarantee that a robot's "blind" manipulation will succeed in moving an object to a specific goal.
Results and The Herding Strategy
Using these analytical bounds, the authors developed a "herding" strategy. By moving a robot finger in a specific spiral path, a disk can be guided toward a goal location without any visual or tactile feedback. The spiral converges toward the goal at a rate of approximately $n^{-1.6}$, where $n$ is the number of revolutions.
The paper also identifies a fundamental "localization limit" imposed by friction ($\mu_c$) between the pusher and the object. They proved that a disk can only be perfectly localized if friction is zero. If friction exists, the disk can only be localized to a circle of radius $b_\infty = a(\mu_c + 1)$. For instance, with a friction coefficient of 0.25, the disk can only be "herded" into a circle 1.25 times its own radius.
Conclusion
By providing a quantitative method to bound the unpredictable nature of sliding, this work laid the foundation for robust, sensorless robot manipulation. While the bounding disk method can be conservative for highly non-circular shapes, it guarantees success by accounting for the worst-case support distributions, proving that friction is the ultimate limit to positioning precision.
Summary updated