Create your own virtual creature that learns via differentiable physics. Use arrow keys to control it.
Grid
Delete
Box
Act

Parameters

Stifness

Actuation

Train Iteration

Learning Rate

Loss Curve

Introduction

This is a mass spring physical simulator integrated differentiable physics. In the canvas, you can draw your own unique robot and train it through neural networks to let it learn to walk(if the structure of your robot allows so).
Alter Grid to draw with or without supporting grid in the canvas.
Alter Delete(delete existing springs), Box(draw a pre-defined box directly), and Act(let the springs you choose move or choose the point to be the head) to switch between the drawing mode.
Click on Act All to let all springs move or not move.
You can also save and load the robot you have drawn to try out different parameters.
Finally, click on Train to see what happens!!!

About the author

Jiasheng Zhang, CS bachelor degree at University of Wisconsin-Madison.
E-Mail: leonz5288@hotmail.com