Deformable-Modeling

A semi-empirical method for simulating contact with elastically deformable objects whose force response is learned using entirely data-driven models.

System Overview

A point interaction model of a deformable object is first learned through data acquired with the system shown here after a small number of pokes. The model is then used in a semi-empircal simulator that solves for the deformation and contact wrench with an arbitrary rigid object. The color of the deformed object represents the amount of surface deformation.

Getting started

Requirements

Hardware:
- Robots with with force sensors (e.g. UR5e)
- A computer with a dedicated GPU(recommended)
- RGB-D Camera (e.g. Intel SR300)
- Probes
OS: Ubuntu 16.04
Environment:
- Python 2.7 / Python 3.5
- Klamp't
- Open3D

Installation

Simply git clone the project and check all the dependent libraries.

Calibration

First, do camera calibration to make the system work correctly. You can do all when connecting the robot.

cd experiment
python calibration_testing.py

cd utils
python calibration_calculation.py
python generate_marker_transform_color.py

then connect robot

python take_calibration_pictures

Executing

simply run

python main.py --process [process-name]

or (recommend)

checkconfig.py, set self.use_mkdir_exp = True , then

python main.py --process debugging --step y --probe point

Version

2019.07.16 First version.
2019.07.19 Add line probe mode, add transform from probe to EE, fix unit vector bug.
2019.07.21 Add duplicates removing, collision detection, and fix point cloud order.
2019.07.23 Change line poking process: first reconstruct a probe list then poke. fix some vis bugs.
2019.07.24 Speed up getting pcd, test all process in physical, update data format, add future folder (2 camera in thread)
2019.08.13 Update the final version and data set.
2020.07.05 Update analysis, models, simulation, and data. Delete future foler. Reagrrange codes.

kailucs/Deformable-Modeling