Let's Push Things Forwards
As robots make their way out of factories to work alongside humans, it is crucial that they develop the necessary skills to manipulate and interact with their environment in various and unforeseeable circumstances. Pushing, with this regards, becomes a key primitive manipulation skill for a robot's repertoire. Imagine a humanoid robot designate to assist elders in their houses, tasked to fetch a medicine from the top shelf. Instead of picking up each object that obstructs the path, the robot could use gentle pushes to create a corridor to its target. Moving larger and heavier obstacles out of the way is even more essential for mobile robots, especially in environments as extreme as Mars, or for rescue missions after an earthquake or tsunami, such as for the Fukushima Daiichi Nuclear Power Plant. Yet highly interesting and entertaining, dexterous pushing skills are also widely applied and applauded in robot soccer.
Robots in warehouses can navigate freely and deliver goods, nevertheless, no robotic system is yet capable of exploiting pushing operations for novel items in novel situations, such as inserting a box of varied produce onto an over-the-head store shelf.
While humans perform skillful manipulation tasks from an early age and learn how to transfer behaviours from one object to objects of novel sizes, shapes, and physical properties, for robots, achieving those goals is very challenging. This complexity arises from the fact that physical properties, such as frictional forces, are usually unknown but play a significant role in order to determine the effects of a push. To complicate the problem further, the dynamics of pushing are highly non-linear, with literal tipping points and sensitive to initial conditions. The mechanics of pushing has been investigated since the early 80s, and the large body of work in this field has produced many efficient models for controlling, planning, and predicting the outcome of a push. However, models that generalise to novel objects are scarce, highlighting the demanding nature of the problem.
Researchers from the University of Birmingham have worked towards the realisation of autonomous robot manipulators that can safely operate everyday objects to assist humans.
Researchers from the University of Birmingham have worked towards the realisation of autonomous robot manipulators that can safely operate everyday objects to assist humans. One major strand of research focused on providing robots with the ability to predict the effects of their own actions onto the world, as when we push something around.
The current state-of-the-art in robot pushing is surveyed for the first time in Jochen Stuben and Claudio Zito's latest paper "Let's Push Things Forward: A Survey on Robot Pushing" published by Frontiers in Robotics and AI journal.
The paper presents a comprehensive and systematic overview of the existing works on the push manipulation, which was sorely needed. The work is primarily targeted at newcomers, such as Ph.D. students, interested in understanding the evolution of the field. The survey categorises an extensive number of articles into six categories: i) purely analytical, ii) hybrid, iii) dynamic analysis, iv) physics engines, v) data-driven, and vi) deep learning. Also, five trends of open problems are identified and discussed: i) understanding and semantic representation, ii) sensor fusion and feedback, iii) explicitly modelling uncertainty in the model, iv) cooperative robots and multiple contact pushing, and v) real-world applications. While the main body of this work focuses on a qualitative analysis of the presented methods, the mathematical treatment is delivered as a set of mini-lectures in the figures, which draw a geometrical intuition for important formalisations used across the literature. Each figure is accompanied by an explanation of the mathematical content in an accessible yet careful way.
While some typical problems still require a better solution, new challenges and requirements are emerging in the field. To make pushing a key skill for robots in practical applications, many challenges will need to be investigated in the future by research group worldwide. Although the theory behind control and motion prediction is well-established, proper industrial applications has not yet been achieved. Robots in warehouses can navigate freely and deliver goods, nevertheless, no robotic system is yet capable of exploiting pushing operations for novel items in novel situations, such as inserting a box of varied produce onto an over-the-head store shelf.