DiffeoMorph: Learning to Morph 3D Shapes Using Differentiable Agent-Based Simulations
Seong Ho Pahng, Guoye Guan, Benjamin Fefferman, Sahand Hormoz · arXiv · 2025
DiffeoMorph is an end-to-end differentiable method that learns agent-based morphogenesis rules to transform minimally patterned starts into target 3D shapes using an invariant Zernike-based shape-matching loss.
Plain English summary
Why this matters
Key findings
- Introduces DiffeoMorph: an end-to-end differentiable framework for learning distributed morphogenesis protocols for 3D shape formation.
- Uses an SE(3)-equivariant graph neural network for agent position and internal-state updates based on local interactions.
- Proposes a 3D Zernike polynomial-based shape-matching loss that compares continuous spatial distributions and is invariant to agent ordering, number of agents, and global orientation.
- Adds an alignment step to rotate the predicted Zernike spectrum to match the target while preserving reflection sensitivity.
- Demonstrates formation of complex shapes from minimally patterned initial conditions and benchmarks the proposed loss against standard metrics.
Limitations
The abstract does not specify experimental validation, physical realization, hardware constraints, robustness to noise, or performance metrics beyond stating that benchmarking and demonstrations were performed.
Publication
- Publisher
- arXiv
- Publication date
- December 18, 2025
- Research type
- Preprint
- arXiv
- 2512.17129
- Access
- open
Tags
More on Programmable Materials
See all →Shape optimization of 4D-printed multi-material morphing structures for enhanced structural stability
This study presents a shape optimization framework for enhancing the stiffness of 4D-printed multi-material morphing structures.
A Versatile‐Designable Framework for Active and Programmable Shape‐Morphing Soft Matter Systems: From Inverse Design to Closed‐Loop Control
This research presents a framework for active and programmable shape-morphing soft matter systems, enhancing soft robotics capabilities.
Architecting three-dimensional reconfigurable matter from pop-up kirigami with programmable multistability
This research presents a new platform for creating programmable multistable pop-up kirigami systems that can transform into complex 3D shapes.
Dimple-Encoded Reprogrammable Origami
This research presents a dimple-encoded origami platform that allows for reprogrammable shape-morphing and adaptive mechanical systems.
Geometric memory in incomplete phase transitions across dimensions
A nucleation-and-growth model with incomplete reversion produces a geometric memory in plate-size distributions, with stronger memory in 2D than in 3D or lamellar geometries.
Logarithmic-Time Geodesically Convex Decomposition in Programmable Matter
It presents an O(log n)-round algorithm to decompose arbitrary amoebot programmable-matter structures into O(|H|) geodesically convex regions using reconfigurable circuits.
Recently published
Latest research →Bio-inspired re-entrant honeycomb metamaterial with programmable dual-plateau mechanical response
A bio-inspired re-entrant honeycomb mechanical metamaterial is described as having a programmable dual-plateau mechanical response.
Modelling the thermo-mechanical responses and shape recovery performance of particle-reinforced shape memory polymer composites in cold and hot programming
A study focused on modeling thermo-mechanical behavior and shape recovery of particle-reinforced shape-memory polymer composites when programmed in cold versus hot conditions.
Functional Shape Recovery Response of Heat-Treated NiTiCu SHAPE Memory Alloy Wire
Heat treatment of NiTiCu shape memory alloy wires enhances their shape recovery response, achieving a high recovery ratio.
Dorsiventrally Bicolored Leaf‐Inspired Metamaterial Absorbers for Tailorable Electromagnetic Absorption
This research proposes a bioinspired metamaterial that achieves high-performance, tailorable electromagnetic absorption across a broad frequency range.