def _is_collision_free(self, gcc, item_name, graspable, grasp, pose):
        """
        Makes sure a particular graspable is collision free
        """
        gripper_pose = [pose[0][0], pose[0][1], GRIPPER_Z_POS]

        rot_obj = autolab_core.RigidTransform.rotation_from_quaternion(pose[1])
        rot_grip = autolab_core.RigidTransform.rotation_from_quaternion(
            self.gripper_rotation)
        world_to_obj = autolab_core.RigidTransform(rot_obj, pose[0], 'world',
                                                   'obj')

        world_to_grip = autolab_core.RigidTransform(rot_grip, gripper_pose,
                                                    'world', 'gripper')
        obj_to_world = world_to_obj.inverse()
        obj_to_grip = world_to_grip.dot(obj_to_world)
        # now have RigidTransform of gripper wrt object, so can pass to collision check

        gcc.set_graspable_object(graspable)
        grasp_collide = gcc.grasp_in_collision(obj_to_grip.inverse(),
                                               item_name)
        approach_collide = gcc.collides_along_approach(grasp,
                                                       self.cc_approach_dist,
                                                       self.cc_delta_approach,
                                                       item_name)

        if grasp_collide or approach_collide:
            #print "collision detected " + item_name
            return False

        return True
Esempio n. 2
0
def grasp_to_world(objectInWorld, graspToObj):
    rot_obj = autolab_core.RigidTransform.rotation_from_quaternion(
        objectInWorld[1])
    world_to_obj = autolab_core.RigidTransform(rot_obj, objectInWorld[0],
                                               'world', 'obj')
    obj_to_world = world_to_obj.inverse()
    g_to_w = obj_to_world.dot(graspToObj)
    return g_to_w
    def _create_grasp_collision_checker(self, state):
        gcc = dexnet.grasping.GraspCollisionChecker(self.gripper)

        # Add all objects to the world frame
        item_names = state['item_names']
        poses = state['poses']
        graspables = {}
        for item_id, pose in poses.items():
            graspable = self.dn.dataset.graspable(item_names[item_id])
            rot_obj = autolab_core.RigidTransform.rotation_from_quaternion(
                pose[1])
            world_to_obj = autolab_core.RigidTransform(rot_obj, pose[0],
                                                       'world', 'obj')
            obj_to_world = world_to_obj.inverse()
            gcc.add_graspable_object(graspable, obj_to_world)

            graspables[item_id] = graspable

        return gcc, graspables
Esempio n. 4
0
from kin_func_skeleton import rotation_3d


####


if __name__ == '__main__':

    ## Defining the transform between the box and the camera ##

    rotation = np.array([[1,       0,       0], # 'random' values, they will need to be changed for the final setup
                         [0, -0.1979,  0.9802],
                         [0, -0.9802, -0.1979]])
    translation = np.array([0.2, -0.8, 0.4])
    rigid_transfo = autolab_core.RigidTransform(rotation=rotation, translation=translation, from_frame='camera', to_frame='world')

    pawn = trimesh.exchange.load.load('pawn.obj')
    pawn.fix_normals()

    xs = []
    ys = []
    zs = []

    ## Plot the object in 3D ##

    for vertex in pawn.vertices:
        pos_in_world = np.matmul(rigid_transfo.matrix, np.array([vertex[0], vertex[1], vertex[2], 1]))

        xs.append(pos_in_world[0])
        ys.append(pos_in_world[1])