For pushing force visulization, add the following code to IsaacGym_Preview_4_Package/isaacgym/python/isaacgym/gymutil.py
class WireframeTriangleGeometry(LineGeometry):
def __init__(self, xdim=0.5, ydim=0.5, pose=None, color=None):
if color is None:
color = (1, 0, 0)
num_lines = 3*2
x = 0.006 * xdim
y = 0.006 * ydim
verts = np.empty((num_lines, 2), gymapi.Vec3.dtype)
verts[0][0] = (0, 0, 0)
verts[0][1] = (0.1, -y, 0)
verts[1][0] = (0, 0, 0)
verts[1][1] = (-0.1, -y, 0)
verts[2][0] = (0.1, -y, 0)
verts[2][1] = (-0.1, -y, 0)
verts[3][0] = (0, 0, 0)
verts[3][1] = (-x, 0.1, 0)
verts[4][0] = (0, 0, 0)
verts[4][1] = (-x, -0.1, 0)
verts[5][0] = (-x, 0.1, 0)
verts[5][1] = (-x, -0.1, 0)
# verts[0][0] = (0, 0, 0)
# verts[0][1] = (-x, 0, 0)
# verts[1][0] = (0, 0, 0)
# verts[1][1] = (0, -y, 0)
if pose is None:
self.verts = verts
else:
self.verts = pose.transform_points(verts)
colors = np.empty(num_lines, gymapi.Vec3.dtype)
colors.fill(color)
self._colors = colors
def vertices(self):
return self.verts
def colors(self):
return self._colors
a1wxv
- grount truth vision env using sampled height map points
a1wxd
- depth image as the vision input, only for RMA student policy learning
python scripts/train_vision.pt --task a1wxv --headless
for teacher policy
python scripts/train_vision.pt --task a1wxd --headless
for student policy, make sure in a1wxd.config
, estimator = True, Load = 'teacher policy path', resume = True
Install realsen on Nvidia Jetson: https://lieuzhenghong.com/how_to_install_librealsense_on_the_jetson_nx/
reach policy
is for arm only
teacher_policy
is IG policy for both arm and dog
dog only
is to only mimicing dog actions but the policy is trained for a dog with an arm
prior
is to mimicking dog actions for a raisim policy
ig dog only
is to mimicing dog action for a IG policy but the policy is only trained with a dog