In this work, we investigate the question of gait discovery for deformable characters without skeletons. Without skeletons, creatures must rely on volume preservation and muscle contractions for control. We model muscle contractions using combination of both isotropic and non-isotropic expansions of deformable voxels. Each muscle voxel expands and contracts following a sinusoidal schedule which are optimized using CMA-ES to maximize forward movement.
As of currently, only windows machines are supported. Please open the project with Visual Studio 2017 and run the project.
Simulator settings which includes material properties, body plans, muscle oscillation schedules, and more are prop.vxl.json. The read version is a modified version of the original JSON format with several additions. Please refer below for more information.
The program have three modes of operation: Display, Replay, Optimize. Controls are mostly set near the top of main.cpp after the imports.
Display runs the simulation according to the JSON configuration. It can be set to record voxel configurations for replay. The specified recording rate can be set using frameTime and the default time at which a recording is generated can be set with recordTime and the save location is set using saveFile.
Replay replays the voxel configurations read in from the specified file. The playback speed can be set using msec_replay. The read location is set using loadFile.
- AD to move camera along X axis
- WS to move camera along Y axis
- QE to move camera along Z axis
- R to reset to beginning
- P to pause simulation/replay
- U to instantly save replay.
Additional optimization controls is set in cmaesNewLibTest.cpp for the number of parameters. The initial parameters and hyperparameters for CMA-ES are also set in cmaesNewLibTest.cpp. By default, every time a new record is achieved, a JSON file containing that configuration is saved in the data directory. One can change the name of the saved file on line 85.
A utility called convertCoords.py in softbody_util allows users to design body plans in VoxCAD and export them into the simulation.
Usage: python convertCoords.py inFile.txt outFile.txt
To obtain inFile.txt, create your design in VoxCAD. Set Lattice Dim under the Workspace tab to 1000mm and export the file as Voxel Coordinates from File -> Export -> Voxel Coordinates. Then, put the export file into the softbody_util directory and input its name in place of inFile.txt.
- libcmaes compiled for Windows using Cygwin
- Eigen
- Glew and FreeGlut
- Forked version of Voxelyze with non-isotropic expansions/contractions
One have to manually set additional include/linker/library directories as needed for the VS2017 project.
The following are additions to the JSON format
Controls how parameters are loaded into the simulation.
| Properties | Description |
|---|---|
| numMaterials | Number of muscle activations to load parameters for. |
| optAmp | Whether to load amplitude information |
| optFreq | Whether to load frequency information |
| optOff | Whether to load offset information |
Under materials, controls which axis the material will mainly expand towards. Axis number of {-1, 0, 1, 2} correspond with {uniform, X, Y, Z} axis expansions.
Controls how muscles are activated according to the sin function. Controls temperature per material to obtain expansions/contractions. One can set the amplitude, frequency, and offset for the activation.
materialNames sets the correspondence between materials and the oscillation. It is possible to set multiple materials to be controlled by the same oscillation/activation. To use activations, one MUST set the names for materials correspondingly.