Use panda as the platform to conduct grasp tasks.
-
The object is in ./egad_eval_set. Bottle with texture.
-
In order to obtain touch & audio sequences, need to train the touch & audio network for certain object. Please refer to https://github.com/rhgao/ObjectFolder.
-
Put the pre-trained network (.pth file) to ./demo folder.
We detect the contact points between the robot arm and gripper. As for the contact point, we chose the geometry center of contact points to
render tactile iimages. And accumulate the normal and shear forces of those contact points to obatin forces.
-
Install Pybullet and the environment for Object folder 2.0.
-
run "python main_sim_grasp_urdf.py": It will show a grasping demo with tactile image rendered in real-time at the same time.
-
Two work mode in "panda_sim_grasp.py": One is grasping the object vertically (step_grasping) , the other is to rotation with relatively fixed
position.(step_rotation)
- run "python main_sim_grasp_test.py": This is a pythoy file to show how the predicted action sequences work o nthe robot arm. The trail file
is in ./test. For each simulation step, read actions from the file and conduct it on the robot arm. (This is used to validate the results of MPC)
- Touch, audio, grasping images and endpose&action sequences are available with "python main_sim_grasp_urdf.py". They coulde be found in ./
results.
- ./gif_gt/num2gif.py is a tool to generate .gif file from image sequences.
This is a special Panda robot with a gelsight sensor on the left finger. We add gelsight to panda in ./franka_panda/panda_1.urdf
Currently, have finished fixed the gelsight onto Panda. And we set the center of gelsight to be the endpose of the whole arm so taht we could control the position of the center of Gelsight and the rotaion of the endpose is along the center of the Gelsight.
TODO:
Fixed the rendering of Taxim well with the gelsight on Panda.
-
Install Pybullet and the environment for Object folder 2.0.
-
run "python Gelsight.py" to start up Pybullet envrioment.
-
Then you'll see the gripper with gelsight rotating along the center of the Gelsight.
-
TODO: Rendering the tactile image with Taxim at the same time.
Use panda as the platform to conduct grasp tasks.
- python localize.py --num N --start M
In this file, panda robot points to certion vertex on the object along the normal vector. Args: --num: the serial number of object --start: the serial number of data in the npy files
- python localize_new.py --num N
Directly generate images for certain vertex along normal vector
Args: --num: the serial number of object
All objects are in ./ObjectFolder/. For each object, need to define its .urdf file.