Пример #1
0
def sample_edge_point(polygon, radius):
    from misc.numerical import sample_categorical
    edges = zip(polygon, polygon[-1:] + polygon[:-1])
    edge_weights = {
        i: max(length(v2 - v1) - 2 * radius, 0)
        for i, (v1, v2) in enumerate(edges)
    }
    # TODO: fail if no options
    while True:
        index = sample_categorical(edge_weights)
        v1, v2 = edges[index]
        t = random.uniform(radius, length(v2 - v1) - 2 * radius)
        yield t * normalize(v2 - v1) + v1
Пример #2
0
def get_closest_edge_point(polygon, point):
    edges = zip(polygon, polygon[-1:] + polygon[:-1])
    best = None
    for v1, v2 in edges:
        proj = (v2 - v1)[:2].dot((point - v1)[:2])
        if proj <= 0:
            closest = v1
        elif length((v2 - v1)[:2]) <= proj:
            closest = v2
        else:
            closest = proj * normalize((v2 - v1))
        if (best is None) or (length((point - closest)[:2]) < length(
            (point - best)[:2])):
            best = closest
    return best
Пример #3
0
def workspace_motion_plan(
        base_manip,
        manipulator,
        vector,
        steps=10):  # TODO - use IK to check if even possibly valid
    distance, direction = length(vector), normalize(vector)
    step_length = distance / steps
    with base_manip.robot:
        base_manip.robot.SetActiveManipulator(manipulator)
        base_manip.robot.SetActiveDOFs(manipulator.GetArmIndices())
        with collision_saver(base_manip.robot.GetEnv(),
                             openravepy_int.CollisionOptions.ActiveDOFs):
            try:  # TODO - Bug in specifying minsteps. Need to specify at least 2 times the desired otherwise it stops early
                traj = base_manip.MoveHandStraight(direction,
                                                   minsteps=10 * steps,
                                                   maxsteps=steps,
                                                   steplength=step_length,
                                                   ignorefirstcollision=None,
                                                   starteematrix=None,
                                                   greedysearch=True,
                                                   execute=False,
                                                   outputtraj=None,
                                                   maxdeviationangle=None,
                                                   planner=None,
                                                   outputtrajobj=True)
                return list(sample_manipulator_trajectory(manipulator, traj))
            except planning_error:
                return None
Пример #4
0
def sample_edge_pose(surface, mesh):
    world_from_surface = trans_from_pose(surface.pose)
    radius = max(length(v[:2]) for v in mesh.vertices)
    origin_from_base = trans_from_point(0, 0, np.min(mesh.vertices[:, 2]))
    for point in sample_edge_point(surface.convex_hull, radius):
        theta = random.uniform(0, 2 * np.pi)
        surface_from_origin = trans_from_quat_point(quat_from_z_rot(theta),
                                                    point)
        yield pose_from_trans(
            world_from_surface.dot(surface_from_origin).dot(origin_from_base))
def update_belief(robot, world, belief, plan):
    # TODO: before or after movement?
    holding, surfaces, items = belief.holding, belief.surfaces, belief.items
    surface_types = {s.type: s for s in world.surfaces}
    item_types = {i.type: i for i in world.items}
    head = robot.GetManipulator('head')
    for action, args in plan:
        if action.name == 'scan_room':
            # TODO: observe objects under some conditions
            surfaces = world.surfaces[:]
        elif action.name == 'move_head':
            #items = world.items[:]
            # TODO: should this be a sense action?
            pass
        elif action.name == 'move_base':
            items = [Object(ty, p, False) for ty, p, _ in items]
        elif action.name == 'register':
            pass
        elif action.name == 'pick':
            if holding is not None: print holding
            assert holding is None
            _, p, g, bg, _ = args # TODO: use g.value
            item = items.pop(get_pose_index(items, p.value))
            holding = Object(item.type, g, False)
        elif action.name == 'place':
            assert holding is not None
            _, p, g, bg, _ = args
            items.append(Object(holding.type, p.value, False))
            holding = None
        else:
            raise ValueError(action.name)
        # TODO: add noise to these?
        #for name, pose in observe_env(robot, p_hit_vis=0.25).items():
        visible = observe_env(robot, p_hit_vis=1)
        print 'Visible:', visible.keys()
        for name, pose in visible.items():
            ty, _ = name.split('-')
            if ty in item_types:
                # TODO: probabilities for detect and register
                for index, item in reversed(list(enumerate(items))):
                    if ty == item.type:
                        point = point_from_pose(item.pose) if is_oriented(item.pose) else item.pose
                        if (point is None) or np.allclose(point_from_pose(pose), point):
                            items.pop(index)
                #index = get_point_index(items, pose)
                distance = length(get_point(head)[:2] - point_from_pose(pose)[:2])
                pose = pose if (distance <= MAX_REG_DISTANCE) else point_from_pose(pose)
                items.append(Object(ty, pose, is_oriented(pose)))
            elif ty in surface_types:
                [matched] = [s for s in surfaces if (s.type == ty) and (s.pose is not None) and
                            np.allclose(s.pose, pose)]
                matched.observations += 1
    return Belief(holding, surfaces, items)
Пример #6
0
def workspace_motion_plan(base_manip, manipulator, vector, steps=10):
    distance, direction = length(vector), normalize(vector)
    step_length = distance / steps
    with base_manip.robot:
        base_manip.robot.SetActiveManipulator(manipulator)
        base_manip.robot.SetActiveDOFs(manipulator.GetArmIndices())
        with collision_saver(base_manip.robot.GetEnv(),
                             openravepy_int.CollisionOptions.ActiveDOFs):
            try:
                traj = base_manip.MoveHandStraight(direction,
                                                   minsteps=10 * steps,
                                                   maxsteps=steps,
                                                   steplength=step_length,
                                                   ignorefirstcollision=None,
                                                   starteematrix=None,
                                                   greedysearch=True,
                                                   execute=False,
                                                   outputtraj=None,
                                                   maxdeviationangle=None,
                                                   planner=None,
                                                   outputtrajobj=True)
                return list(sample_manipulator_trajectory(manipulator, traj))
            except planning_error:
                return None
Пример #7
0
def get_mesh_radius(body):
    [geometry] = get_geometries(body)
    mesh = geometry.GetCollisionMesh()
    return max(length(v[:2]) for v in mesh.vertices)
    def sample_grasp_traj(obj, pose, pose2):
        enable_all(bodies, False)
        body = bodies[obj]
        #body.Enable(True) # TODO: fix this
        set_pose(body, pose.value)

        # TODO: check if on the same surface

        point = point_from_pose(pose.value)
        point2 = point_from_pose(pose2.value)
        distance = length(point2 - point)
        if (distance <= 0.0) or (0.6 <= distance):
            return
        direction = normalize(point2 - point)
        orientation = quat_from_angle_vector(angle_from_vector(direction),
                                             [0, 0, 1])

        #rotate_direction = trans_from_quat_point(orientation, unit_point())

        steps = int(math.ceil(distance / PUSH_MAX_DISTANCE) + 1)
        distances = np.linspace(0., distance, steps)
        points = [point + d * direction for d in distances]
        poses = [pose_from_quat_point(orientation, point) for point in points]
        pose_objects = [pose] + map(Pose, poses[1:-1]) + [pose2]
        print distance, steps, distances, len(poses)

        radius, height = get_mesh_radius(body), get_mesh_height(body)
        contact = cylinder_contact(radius, height)

        pushes = []
        # TODO: I could do all trajectories after all pushes planned
        for i in xrange(len(poses) - 1):
            p1, p2 = poses[i:i + 2]
            # TODO: choose midpoint for base
            ir_world_from_gripper = manip_from_pose_grasp(p1, contact)
            set_manipulator_conf(arm, carry_arm_conf)
            for world_from_base in islice(
                    random_inverse_reachability(ir_model,
                                                ir_world_from_gripper),
                    max_failures):
                set_trans(robot, world_from_base)
                set_manipulator_conf(arm, carry_arm_conf)
                if check_collision(robot):
                    continue
                q = Conf(robot.GetConfigurationValues())
                push_arm_confs = []
                approach_paths = []
                for p in (p1, p2):
                    # TODO: make sure I have the +x push conf
                    world_from_gripper = manip_from_pose_grasp(p, contact)
                    set_manipulator_conf(arm, carry_arm_conf)
                    grasp_arm_conf = solve_inverse_kinematics(
                        arm, world_from_gripper, collisions=False)
                    if grasp_arm_conf is None:
                        break
                    push_arm_confs.append(grasp_arm_conf)
                    set_manipulator_conf(arm, grasp_arm_conf)
                    pregrasp_arm_conf = solve_inverse_kinematics(
                        arm,
                        world_from_gripper.dot(gripper_from_pregrasp),
                        collisions=False)
                    if pregrasp_arm_conf is None:
                        break
                    #if DISABLE_MOTIONS:
                    if True:
                        approach_paths.append([
                            carry_arm_conf, pregrasp_arm_conf, grasp_arm_conf
                        ])
                        continue
                    """
          set_manipulator_conf(arm, pregrasp_arm_conf)
          grasp_path = plan_straight_path(robot, grasp_arm_conf, pregrasp_arm_conf)
          # robot.Release(body)
          if grasp_path is None:
            continue
          pregrasp_path = plan_path(base_manip, pregrasp_arm_conf, carry_arm_conf)
          if pregrasp_path is None:
            continue
          t = Prehensile(full_from_active(robot, grasp_path + pregrasp_path[1:]), obj, pose, grasp)
          yield [(q, t)]
          reset_env()
          return
          """
                else:
                    # Start and end may have different orientations
                    pq1, pq2 = push_arm_confs
                    push_path = plan_straight_path(robot, pq1, pq2)
                    # TODO: make sure the straight path is actually straight
                    if push_path is None:
                        continue
                    po1, po2 = pose_objects[i:i + 2]
                    ap1, ap2 = approach_paths
                    m = Push(
                        full_from_active(robot, ap1),
                        full_from_active(robot, push_arm_confs),
                        #full_from_active(robot, push_path),
                        full_from_active(robot, ap2[::-1]),
                        obj,
                        po1,
                        po2,
                        contact)
                    pushes.append(PushMotion(obj, po1, po2, q, m))
                    break
            else:
                print 'Failure', len(pushes)
                return
        print 'Success', len(pushes)
        yield pushes