Developed a physics based digital twin of a soft spherical robot that models URDF based rigid body dynamics with pressure dependent hydroelastic contact and joint friction models, then validates the resulting behavior against physical experiments across multiple inflation pressures.
Published in prestigious IEEE Robotics and Automation Journal
IEEE Paper
Average speed error vs. rigid point contact
~88% ↓
Pressure frequency regression
R² = 0.972
Damping ratio error after tuning
67.7% ↓
Contact representation
Point / rigid contact
Single contact pointConcentrated force
Key differences
Contact area
LocalizedFinite patch
Force model
Point forceDistributed load
Rolling realism
Rolls too freelyResists rolling
Hydroelastic contact
Finite contact patchDistributed pressure field
Contact modelForce responseRobot motion
01Problem
Point contact dynamics could not fully capture the robot's motion because the deformable shell generates pressure dependent contact forces that are absent in a point contact model.
Standard point contact models treated the shell as rigid, missing the experimentally observed pressure dependent frequency and damping behavior.
Joint losses also affected the robot's motion, requiring friction models beyond the proportional damping available natively through the URDF representation.
Without modeling these non-conservative effects, simulations could significantly over or under predict the drive velocity measured on the physical robot.
02Approach
Built a modular dynamics model that combines URDF based rigid body mechanics with realistic hydroelastic contact dynamics and joint friction models to reproduce robotic ball's real motion.
Created custom URDF from a SolidWorks assembly, preserving measured mass properties, inertias, link geometry, and the robot's structure.
Isolated the shell dynamics experimentally and fit pressure dependent frequency and damping relationships, then converted them into additional stiffness and damping terms.
Compared rigid point contact, injected shell dynamics, and hydroelastic contact modeling to validate which approach was the most realistic.
03Result
The contact model substantially improved dynamics with the hardware and showed that pressure dependent hydroelastic mechanics were necessary.
The pressure to natural frequency regression reached R² = 0.972, providing a strong empirical relationship for parameterizing shell behavior across inflation pressures.
Tuning the hydroelastic dissipation coefficient cut the error between simulated and experimentally measured shell damping ratios by 67.7%, providing the contact parameters used for final validation.
Across multiple hardware validation tests, the hydroelastic contact model reduced average sim to hardware drive speed error from ~49% with rigid point contact to ~6%, an ~88% reduction.