Yeonsu Jung
jung [at] seas.harvard.edu
Snake Lattice: Inverse Design of Curve Networks
What happens when the robot is not a single chain, but a whole network? Can we design (or control) the topology of an elastic lattice?
From chains to networks
The snake-control project asks how to control a single hyper-redundant chain. Snake lattice asks a harder question: what if the “robot” is a network of interconnected elastic rods — a grid, a lattice, or an arbitrary graph?
This problem arises naturally in:
- Woven and knitted textiles (where filaments intersect at many nodes)
- Elastic metamaterials (where the network topology determines mechanical response)
- Soft robotic skins and grippers with distributed actuation
What the code does
Snake lattice is built around the Discrete Mechanics (DisMech) framework, and this project is one of DisMech’s canonical examples. It simulates and analyzes curve networks with:
- JAX-based automatic differentiation for computing elastic energies and gradients
- Polyscope for 3D visualization of network configurations
- Topology analysis: tracking the graph structure (nodes, edges, edge pairs) as the network deforms
Configurations are stored as .npy files containing node positions and edge connectivity, making it easy to save, reload, and compare configurations.
The analysis scripts cover:
analysis.py— In-plane deformation and force responseanalysis_out_of_plane.py— Out-of-plane buckling and topology changes
Inverse design perspective
The forward problem is: given a network topology and applied forces/displacements, what is the equilibrium configuration? This is standard structural mechanics.
The inverse problem is: given a desired shape or mechanical response, what network topology and rest-state configuration achieves it? This is a design problem. By differentiating through the elastic energy with JAX, we can compute gradients with respect to node positions, rest lengths, or cross-sectional properties, and use gradient descent to find designs that meet a target.
Out-of-plane behavior
A particularly interesting regime is when the network buckles out of plane. In-plane deformations are relatively tractable, but out-of-plane buckling introduces new topological degrees of freedom: overcrossings and undercrossings that change the linking structure of the network. Analyzing these transitions is one of the goals of this project.
| back to works | home |