Learning to Slide Unknown Objects with Differentiable Physics Simulations

Abstract

We propose a new technique for pushing an unknown object from an initial configuration to a goal configuration with stability constraints. The proposed method leverages recent progress in differentiable physics models to learn unknown mechanical properties of pushed objects, such as their distributions of mass and coefficients of friction. The proposed learning technique computes the gradient of the distance between predicted poses of objects and their actual observed poses, and utilizes that gradient to search for values of the mechanical properties that reduce the reality gap. The proposed approach is also utilized to optimize a policy to efficiently push an object toward the desired goal configuration. Experiments with real objects using a real robot to gather data show that the proposed approach can identify mechanical properties of heterogeneous objects from a small number of pushing actions.

Fast and Feature-Complete Differentiable Physics Engine for Articulated Rigid Bodies with Contact Constraints

“…Earlier work computed symbolic gradients for smooth dynamic systems [25], and [7] simplified the computation of the derivative of the forward dynamics exploiting the derivative of the inverse dynamics. Symbolic differentiation becomes manageable when the gradients are only required within smooth contact modes [42] or a specific contact mode is assumed [38]. Recently, Amos and Kolter proposed a method, Opt-Net, that back-propagates through the solution of an optimization problem to its input parameters [2].…”

A Review of Physics Simulators for Robotic Applications

“…The primary benefit of differentiable simulation -the ability to use gradient based rather than black-box optimisation approaches -promises a leap in efficiency and opens up previously intractable problems to learning-based optimisation. Several papers have proven examples which show the applicability of such simulators for system ID [153], policy creation [154] and embedding physics in neural networks (Fig. 4) [155].…”

Differentiable Simulation for Physical System Identification

“…The field of system identification [23] intends to build a mathematical model of a dynamical system from its measurements. The related work [8], [24], [25], [26] identifies parameters of physical systems using simulators as generative models. In each case, the identification is done by simulating the physical system and then optimizing the physical parameters so that the simulations are fitting to the real scenes.…”

“…(a) (b) Fig. 5: Limitations of simulators approximating the friction cone with a pyramid [8], [11], [24], [26]. On Fig.…”

“…5a. Thus in 3D, because pyramidal cones are not isotropic, a same friction coefficient value may lead to two different simulated trajectories when using formulations based on this kind of approximation [8], [11], [24], [26]. Similarly, in the context of friction estimation, the same observed motion may then lead to two different friction values depending on the orientation of the frictional pyramid (Fig.…”