Learning to Move Objects with Fluid Streams in a Differentiable Simulation

TL;DR

This paper presents a neural network controller trained in a differentiable simulation to manipulate objects in 3D space using fluid streams, enabling contactless object movement with only object state observations.

Contribution

It introduces a novel differentiable simulation framework for fluid-object interactions and demonstrates effective control of objects using fluid streams with limited training iterations.

Findings

Controller generalizes to longer episodes

Learns complex fluid-solid dynamics

Requires only object state observations

Abstract

We introduce a method for manipulating objects in three-dimensional space using controlled fluid streams. To achieve this, we train a neural network controller in a differentiable simulation and evaluate it in a simulated environment consisting of an 8x8 grid of vertical emitters. By carrying out various horizontal displacement tasks such as moving objects to specific positions while reacting to external perturbations, we demonstrate that a controller, trained with a limited number of iterations, can generalise to longer episodes and learn the complex dynamics of fluid-solid interactions. Importantly, our approach requires only the observation of the manipulated object's state, paving the way for the development of physical systems that enable contactless manipulation of objects using air streams.