Soft systems have the potential to dramatically improve several areas of robotic manipulation in space. Soft robots incorporate safety at the material level and are generally biocompatible. They are usually significantly lighter than hard robots: the substrates can be a soft polymer, cloth, or many other materials, instead of the traditional metal and resin structures in hard robots. The systems can be designed to have continuous deformation, and many substrates can be cured and modified in relatively simple lab setups with low energy input requirements. However, these substrates but are notoriously difficult to control. To date, many actuation and locomotion methods have been proposed, but soft robots still lack the ability to interact with their environment in ways similar to traditional robotic manipulators or biological arms and tentacles. I propose a new approach to soft robot control comprising a combination of soft sensor-actuator pairs and machine learning. This foundational method can be applied to the broad class of soft materials, for 3D closed-loop control of arbitrary structures. I will demonstrate the approach’s efficacy on a soft 6-degree-of-freedom (DOF) closed-loop robotic manipulator, closing this key knowledge gap.