More than a Million Ways to Be Pushed: A High-Fidelity Experimental Dataset of Planar Pushing (1604.04038v2)

Abstract: Pushing is a motion primitive useful to handle objects that are too large, too heavy, or too cluttered to be grasped. It is at the core of much of robotic manipulation, in particular when physical interaction is involved. It seems reasonable then to wish for robots to understand how pushed objects move. In reality, however, robots often rely on approximations which yield models that are computable, but also restricted and inaccurate. Just how close are those models? How reasonable are the assumptions they are based on? To help answer these questions, and to get a better experimental understanding of pushing, we present a comprehensive and high-fidelity dataset of planar pushing experiments. The dataset contains timestamped poses of a circular pusher and a pushed object, as well as forces at the interaction.We vary the push interaction in 6 dimensions: surface material, shape of the pushed object, contact position, pushing direction, pushing speed, and pushing acceleration. An industrial robot automates the data capturing along precisely controlled position-velocity-acceleration trajectories of the pusher, which give dense samples of positions and forces of uniform quality. We finish the paper by characterizing the variability of friction, and evaluating the most common assumptions and simplifications made by models of frictional pushing in robotics.

• The paper presents a comprehensive, high-fidelity experimental dataset capturing over a million planar pushing interactions across six varied dimensions to support robotic manipulation research.
• Key findings highlight significant variability in dynamic surface friction based on location, time, speed, and direction, revealing limitations of traditional constant-coefficient friction models like Coulomb.
• This dataset serves as a crucial resource for validating existing predictive models and developing new data-driven approaches that better accommodate stochastic friction in robotic manipulation tasks.

High-Fidelity Dataset of Planar Pushing for Robotic Manipulation

The paper "More than a Million Ways to Be Pushed: A High-Fidelity Experimental Dataset of Planar Pushing," authored by Yu et al., presents a comprehensive dataset aimed at elucidating the mechanics of planar pushing—a fundamental motion primitive useful in robotic manipulation. The researchers at Massachusetts Institute of Technology have addressed the inherent complexities in predicting and understanding the motion of objects subjected to pushing interactions. They propose to bridge the gap between theoretical assumptions in robotics and empirical data by offering a dataset that spans several critical dimensions of planar pushing.

In robotic manipulation, pushing serves multiple functions: facilitating reorientation, aiding perceptions, or assisting in grasping large or cluttered objects. Though theoretical models of planar pushing—typically grounded on Coulomb's friction principles—exist, they often result in oversimplified deterministic systems that may not effectively capture real-world variability. Critical to advancing this field is the introduction of empirical datasets against which predictive models can be measured, validated, and refined.

The dataset introduced captures detailed information about the poses and forces involved in pushing interactions, automated to ensure precision and reproducibility. Specifically, it varies across six dimensions:

1. Surface Material
2. Shape of the Pushed Object
3. Contact Position
4. Pushing Direction
5. Pushing Speed
6. Pushing Acceleration

The authors employed an industrial robot equipped with a high-precision force torque sensor and a Vicon tracking system, ensuring data fidelity through automation.

Key Findings and Implications

The paper provides insights into the variability of dynamic surface friction with respect to location, time, speed, and direction. Tests revealed that surface imperfections crucially influence the effective dynamic coefficient of friction (DCoF). Notably, the variability is not consistent across materials like ABS, Delrin, plywood, and polyurethane—highlighting the limitations of traditional Coulomb and quasistatic friction assumptions.

Further, the dataset evaluates the principle of maximum-power inequality and the ellipsoidal approximation of limit surfaces. It was found that, while some assumptions remain contextually valid, the coefficient of friction varied significantly in practical scenarios. Such findings encourage the development of new models that might better accommodate the stochastic nature of friction observed in experiments.

This dataset emerges as a powerful asset not only for validation of existing models but also for inspiring novel approaches to handling variability and uncertainty in robotic manipulation. Such data-driven insights may contribute to refining planning algorithms, enhancing control strategies, and advancing simulations that accommodate real-world complexities.

Future Directions in Robotic Manipulation

Analyzing finer aspects of non-linear interactions in manipulation tasks—especially when faced with uncertainty—could benefit from datasets like the one introduced in this paper. Progressing towards more sophisticated and adaptable models of friction could lead to enhanced robotic autonomy—moving beyond open-loop planning to strategies accounting for the stochastic nature of real-world interaction forces.

The dataset thus serves as a foundational tool, potentially catalyzing future research aimed at modeling non-prehensile interactions with greater accuracy. This aligns with long-term goals of improving robotic systems by incorporating more reliable feedback mechanisms and adaptable planning strategies, ensuring robots can act with increased precision and autonomy when faced with environmental uncertainties.

Yu et al. effectively pave the way for advancements in robotics through empirical validation and modeling, impacting both theoretical insights and practical applications in the field of robotic manipulation.