Origami-Inspired Reconfigurable Robot

A four-legged robot built from origami-inspired modules whose joints switch between soft and rigid, so the same legs can walk, crawl, inch, or grasp.

The four-legged reconfigurable robot built from origami-inspired modules

A frog walks, jumps, swims, and climbs with the same body and legs, just held in different configurations. Most robots, in contrast, need a separate mechanism for each way of moving. This robot takes the frog’s approach: its legs are stacks of small origami-inspired modules whose joints can each be made soft or rigid. Pulling the same tendon then produces a different motion depending on which joints are soft, so the same four legs can walk, crawl, inch, or act as arms to grasp.

At a glance

Feature Detail
Module Two triangular plates (45 mm sides) joined by three vertical links and three diagonal links, driven by a tendon
Joint Shape-morphing variable-stiffness joint: a PLA rod in an elastic tube, softened by a heating wire
Module motions Collapse, bend, or slant under the same tendon pull, depending on which joints are soft
Model Kinematic model predicts the module’s motion with a mean error of 2.4 mm
Robot Four legs, each made of four modules in series
Locomotion Walking (28 mm/s), crawling (21 mm/s), crawling while grasping with two legs (21 mm/s), and inching

The module and its joints

A single origami-inspired module
How a shape-morphing variable-stiffness joint is bent into a new shape and frozen there

Each module is a pair of triangular plates connected by three vertical links and three diagonal links. Each diagonal link has a shape-morphing variable-stiffness joint: a thermoplastic (PLA) rod inside an elastic tube, wrapped with a heating wire. Three volts heat the rod soft in about 20 seconds so the joint can bend; when it cools, the joint becomes rigid again, either straight or frozen in its new bent shape (right).

A single tendon pulls the top plate down. Which joints are soft decides what happens:

All three joints soft: the module collapses flat.
Two joints soft: the module bends.
One joint soft: the module slants.

Stacking modules multiplies the options. Two modules in series can collapse or bend either module, or both, giving a much larger set of shapes and motions from one tendon.

Motions of two modules connected in series.

A robot that walks, crawls, inches, and grasps

The robot has a three-part body and four legs, each made of four modules in series and driven by a tendon. By choosing which joints are rigid in each leg, the same robot moves in different ways:

Walking with upright legs, at 28 mm/s.
Crawling with the body low to the ground, at 21 mm/s.
Crawling while grasping: the front legs hold an object while the rear legs crawl.
Inching between two walls, about 60 mm per cycle.

For these robot demonstrations, the heated joints were replaced by small clips that make a joint soft or hold it rigid, straight or bent, and the legs were reconfigured by hand between modes. Using the heated joints throughout would let the robot reconfigure itself without manual help.

A heated joint replaced by a soft clip or a rigid clip

Follow-up: learning the best stiffness for a task (2026)

In the original robot, we picked which joints to make soft by trial and error. In a follow-up study with collaborators at UMass Amherst, we let a computer decide. Using a two-module, tendon-driven manipulator, a reinforcement learning framework in the MuJoCo physics simulator learns the joint stiffnesses and the tendon control policy together for a reaching task. The learned stiffnesses are built into a physical prototype with fixed-stiffness joints, and the learned controller transfers directly: the real manipulator reaches its targets repeatably, with a final position error below 2 mm.

Co-optimizing stiffness and control in simulation, then transferring to the prototype

Reaching in simulation.
The same reach on the prototype (2× speed).

Team

Zhe Chen, Brandon Tighe, and Jianguo Zhao (Chen et al., 2022). The follow-up was done with Li Chen and Hao Zhang at the University of Massachusetts Amherst (Chen et al., 2026). This work was part of Zhe Chen’s Ph.D. dissertation.

References

2026

  1. T-Mech
    chen2026cooptimizing.jpg
    Co-Optimizing Physical Reconfiguration Parameters and Controllers for a Reconfigurable Manipulator
    Zhe Chen, Li Chen, Hao Zhang, and 1 more author
    IEEE/ASME Transactions on Mechatronics, 2026

2022

  1. T-Mech
    chen2022origami.jpg
    Origami-Inspired Modules Enable A Reconfigurable Robot with Programmable Shapes and Motions
    Zhe Chen, Brandon Tighe, and Jianguo Zhao
    IEEE/ASME Transactions on Mechatronics, 2022