Fast adaptation for effect-aware pushing

Abstract

Abstract-In order to produce robots that are more capable of skilled manipulation tasks, they must learn meaningful knowledge of how objects behave to external stimulus. With this knowledge a robot can predict the outcome of an action, control the object to serve a particular purpose, and together with reasoning, create or modify robot plans. In this paper we 1) build a mathematical compact model for planar sliding motion of an object, 2) show how a robot acquires the parameters of such a model; then how this is used to 3) predict pushing actions; and 4) to move an object from any 1 position and orientation to another.

Learning stable pushing locations

“…Notably, Ruiz-Ugalde et al execute a pushing behavior by determining the static and kinetic friction coefficients, both between the robot hand and object and between the object and table, for multiple objects with rectangular footprints [3]. Additionally they present a robust feedback controller using a cart model for the object being pushed.…”

“…As such there has been considerable work at developing such capability. Early work that analyzed a complete model of the dynamics of pushing was developed by Mason who describes the qualitative rotational changes of sliding rigid objects being pushed by either a single point or single line contact [1]; representative examples of some more recent applications of pushing are available in [3,8,[10][11][12][13][14].…”

“…For an unknown object these properties cannot be deduced without interactive experimentation. Even if such a model is available, simulation may not be sufficient in solving the pushing control problem as efficient solutions require assumptions of uniform, stationary friction coefficients of both the object and the supporting surface [1][2][3].…”

Distinguishing sliding from slipping during object pushing

“…In [4], a visual tracking system was employed to maintain a suitable angle of contact in order to ensure proper friction. However, online slippage monitoring and active force control call for tactile sensors within the feedback loop.…”

“…Knowing the friction coefficient, one can then estimate slippage from the ratio of observed tangential and normal forces at the contact point [7]. In [4], the friction coefficient to be considered during pushing is estimated in advance by sliding over the fixated object and estimating the contact force from measured torques at the hand joints.…”