Robotic arm optimizes silicone 3D printing in gel: Study shows multi-axis toolpaths for nearly full infill

Soft silicone components can also be produced in 3D printing with overhangs if the material is extruded into a supporting gel. This “embedded silicone printing” (ESP) stabilizes the structure during printing. However, typical problems arise with voluminous, almost solidly filled models: silicone and gel are not miscible, filaments fuse less easily, and high overlaps lead to too much material locally. In addition, classic, planar layers on curved surfaces create visible steps.

This is precisely where the study “Robot-assisted multi-axis embedded silicone printing for free-form volumetric model” comes in. In it, Hailin Sun, Yingjun Tian, Chenyu Xu, Mahdi Bodaghi, Fei Gao, and Guoxin Fang describe a multi-axis ESP framework that combines spatial tool paths with a robotic arm. The paper appeared in the journal Virtual and Physical Prototyping and was published online on January 16, 2026 (DOI: 10.1080/17452759.2026.2614811).

The approach uses a curved slicing strategy based on an optimized scalar field. The goal is to achieve a more even distribution of layer heights without losing critical surface features. This is followed by boundary-conforming path planning with staggered paths between layers. This staggering is intended to reduce vertical pores that can occur with dense infill. In addition, the method links the path width to the local layer height. This is intended to better match the amount of material to the respective geometry despite a constant extrusion rate and avoid overfilling or underfilling.

The process was validated on a multi-axis robot platform (UR5e) with mixed extrusion in a carbomer gel bath. Ecoflex 0030 and Dragon Skin 10 and 30 were processed, each with additives to adjust viscosity and reaction behavior. The authors report an infill value of 99.47 percent and surface deviations of less than 1.5 millimeters for test objects from wearables, biomedical phantoms, and soft robotics. In the aortic model, the average error decreases from 0.49 to 0.20 millimeters. An accompanying video is also available.

The entire paper can be read here.