Data-Driven Modeling and Control of Soft Robotic Systems

Overview

This project presents a unified framework for modeling and control of soft robotic systems, combining physics-based modeling with data-driven methods.

The work integrates:

A muscle–rod PDE model for soft actuator dynamics
Koopman operator learning using EDMDc
A control framework based on PIKO (Physics-Informed Koopman Operator)

📄 Paper:
Read on IEEE Xplore

1. Muscle–Rod PDE Model

We model the soft robotic actuator using a continuum mechanics formulation, where the system is described by a set of nonlinear partial differential equations (PDEs).

Key features:

Captures continuous deformation dynamics
Incorporates actuation forces and material properties
Provides a physics-consistent baseline model

This model serves as the foundation for generating simulation data and understanding system behavior.

2. Koopman Operator Learning (EDMDc)

To enable efficient control, we approximate the nonlinear system using a Koopman operator framework.

We use:

EDMDc (Extended Dynamic Mode Decomposition with control)
Learned linear dynamics in a lifted space

Advantages:

Converts nonlinear dynamics → linear representation
Enables model predictive control (MPC)
Efficient for real-time applications

3. PIKO Framework

We propose a Physics-Informed Koopman Operator (PIKO) approach that integrates:

Physics-based structure from the PDE model
Data-driven learning from system trajectories

This hybrid approach improves:

Model accuracy
Generalization across operating conditions
Stability in control tasks

Results

The proposed framework demonstrates:

Accurate prediction of soft robot dynamics
Improved control performance compared to baseline models
Real-time feasibility for robotic applications

Code & Reproducibility