Parallel Simulation of Contact and Actuation for Soft Growing Robots

Soft growing robots, commonly referred to as vine robots, have demonstrated remarkable ability to interact safely and robustly with unstructured and dynamic environments. It is therefore natural to exploit contact with the environment for planning and design optimization tasks. Previous research has focused on planning under contact for passively deforming robots with pre-formed bends. However, adding active steering to these soft growing robots is necessary for successful navigation in more complex environments. To this end, we develop a unified modeling framework that integrates vine robot growth, bending, actuation, and obstacle contact. We extend the beam moment model to include the effects of actuation on kinematics under growth and then use these models to develop a fast parallel simulation framework. We validate our model and simulator with real robot experiments. To showcase the capabilities of our framework, we apply our model in a design optimization task to find designs for vine robots navigating through cluttered environments, identifying designs that minimize the number of required actuators by exploiting environmental contacts. We show the robustness of the designs to environmental and manufacturing uncertainties. Finally, we fabricate an optimized design and successfully deploy it in an obstacle-rich environment.

💡 Research Summary

This paper presents a comprehensive framework that unifies the modeling of growth, bending, pneumatic actuation, and environmental contact for soft growing robots, commonly known as vine robots. The authors extend the thin‑walled inflated‑beam model with a wrinkling‑based restoring moment formulation that accurately captures the small‑angle bending behavior (≤10°) often observed during steering, overcoming the over‑estimation inherent in previous maximum‑wrinkling assumptions. For actuation, they adopt serial pneumatic artificial muscles (sP‑AMs) and refine the existing Pleated Pneumatic Artificial Muscle (PP‑AM) model to include saturation effects based on actuator radius and length. The resulting force‑strain relationship is expressed with incomplete elliptic integrals and calibrated with an empirically determined pressure‑dependent factor.

The robot body is discretized into rigid segments of 25 mm length, a granularity chosen to balance computational cost and model fidelity. For each segment, the net restoring moment is computed as the difference between the wrinkling‑based beam moment and the moment generated by the sP‑AM contraction force, scaled by the effective lever arm (2 R_vine + R_act). This combined moment serves as a constraint in a Position‑Based Dynamics (PBD) solver, which updates the robot’s configuration at each time step while simultaneously handling growth, bending, actuation, and contact forces.

A key contribution is the implementation of this forward dynamics pipeline on GPUs using PyTorch and JAX. The authors replace the costly analytical evaluation of the sP‑AM force‑strain curve with a neural surrogate model trained on a dense dataset generated from the analytical equations, achieving orders‑of‑magnitude speed‑ups. The batch‑processing capability enables thousands of simulations to run in parallel, which is essential for the downstream design optimization task.

In the optimization stage, the authors formulate a black‑box problem: minimize the number of actuators required for a vine robot to navigate a cluttered environment while reaching a target location. They employ a batched version of Stable Sparse RRT (SST*) as the planner, leveraging the parallel simulator to evaluate many control sequences simultaneously. Robustness is incorporated by sampling variations in material properties, manufacturing tolerances, and environmental uncertainties, ensuring that the resulting designs are not overly sensitive to real‑world variations.

Experimental validation includes both simulation‑to‑real comparisons and a physical prototype built from the optimized design. In a series of obstacle‑rich scenarios, the robot’s trajectory matches the simulated prediction within a mean positional error of less than 5 mm. The optimized design achieves a reduction of more than 30 % in the number of required actuators compared to a baseline design, while maintaining a success rate above 95 % across varied trials.

The authors release the entire simulation and optimization code as open source (https://github.com/CoMMALab/ActVineSimPy), facilitating reproducibility and further research. In summary, this work bridges the gap between accurate physical modeling of actively steered soft growing robots and high‑throughput computational tools needed for planning and design, opening new possibilities for deploying vine robots in medical, inspection, and exploration tasks where complex contact‑rich environments are the norm.