示例#1
0
def to_blocks(tower):
    blocks = []
    for ix in range(tower.shape[0]):
        block = Object('block',
                       dimensions=tower[ix, 4:7].numpy().tolist(),
                       mass=tower[ix, 0].item(),
                       com=tower[ix, 1:4].numpy().tolist(),
                       color=(1, 0, 0))
        block.pose = Pose(Position(*tower[ix, 7:10].numpy().tolist()),
                          Quaternion(*tower[ix, 10:14].numpy().tolist()))
        blocks.append(block)
    return blocks
示例#2
0
def main(args):
    NOISE=0.00005

    # define real world block params and initial poses
    #TODO: Load blocks from pickle file.
    if args.use_vision:
        with open(args.blocks_file, 'rb') as handle:
            blocks = pickle.load(handle)
        block_init_xy_poses = None  # These will be initialized randomly but updated by the vision system.
    else:
        # block0 = Object('block0', Dimensions(.0381,.0318,.0635), 1.0, Position(0,0,0), Color(1,0,0))
        # block1 = Object('block1', Dimensions(.0381,.0587,.0635), 1.0, Position(0,0,0), Color(0,0,1))
        # block2 = Object('block2', Dimensions(.0635,.0381,.0746), 1.0, Position(0,0,0), Color(0,1,0))
        block0 = Object('block0', Dimensions(.0381,.0318,.05), 1.0, Position(0,0,0), Color(1,0,0))
        block1 = Object('block1', Dimensions(.0381,.0587,.06), 1.0, Position(0,0,0), Color(0,0,1))
        block2 = Object('block2', Dimensions(.0635,.0381,.05), 1.0, Position(0,0,0), Color(0,1,0))
        blocks = [block0, block1, block2]

        block_init_xy_poses = [Pose(Position(0.65,0.3,0), Quaternion(0,0,0,1)),
                               Pose(Position(0.65,0.15,0), Quaternion(0,0,0,1)),
                               Pose(Position(0.65,0.0,0), Quaternion(0,0,0,1))]

    panda = PandaAgent(blocks, NOISE,
                       use_platform=False,
                       block_init_xy_poses=block_init_xy_poses,
                       teleport=False,
                       use_vision=args.use_vision,
                       use_action_server=args.use_action_server,
                       real=args.real)

    # for now hard-code a tower, but in the future will be supplied from
    # active data collection or tower found through planning for evaluation
    tower_blocks = copy.copy(blocks)
    if args.use_vision:
        tower_poses = [Pose(Position(0.5,-0.25,0.0725), Quaternion(0,0,0,1)),
                        Pose(Position(0.5,-0.25,0.18), Quaternion(0,0,0,1)),
                        Pose(Position(0.5,-0.25,0.28), Quaternion(0,0,0,1))]
    else:
        tower_poses = [Pose(Position(0.3,0.25,.0318), Quaternion(0,0,0,1)),
                        Pose(Position(0.3,0.25,.0953), Quaternion(0,0,0,1)),
                        Pose(Position(0.3,0.25,.1643), Quaternion(0,0,0,1))]

    tower = []
    for block, pose in zip(tower_blocks, tower_poses):
        block.set_pose(pose)
        block = get_rotated_block(block) # NOTE: have to do to set rotations field of block
        tower.append(block)

    # and execute the resulting plan.
    if args.use_action_server:
        panda.simulate_tower_parallel(tower, base_xy=(0.5, -0.25), real=args.real, vis=True, T=2500)
    else:
        panda.simulate_tower(tower, base_xy=(0.5, -0.25), real=args.real, vis=True, T=2500)
示例#3
0
    def __init__(self, direction=None, timesteps=50, rot=None, delta=0.005, block=None):
        """ PushAction moves the hand in the given world by a fixed distance
            every timestep. We assume we will push the block in a direction through
            the object's geometric center.
        :param world: The world which this action should apply to. Used to get the hand
                      and calculate offsets.
        :param direction: A unit vector direction to push.
        :param timesteps: The number of timesteps to execute the action for.
        :param delta: How far to move each timestep.
        """
        super(PushAction, self).__init__(T=timesteps)

        self.rot = rot
        self.timesteps = timesteps
        self.delta = delta
        self.tx = 0

        if direction is None:
            self.direction = self.get_random_dir()
        else:
            self.direction = direction

        _, _, block_height = np.abs(rot.apply(block.dimensions))

        table, leg = Object.platform()
        platform_height = table.dimensions.z + leg.dimensions.z

        self.push_start_pos = Position(x=0 - self.direction[0]*self.delta*20,
                                       y=0 - self.direction[1]*self.delta*20,
                                       z=platform_height + block_height/2 \
                                           - self.direction[2]*self.delta*20)
def plot_constructability_over_time(logger):
    tower_keys = ['2block', '3block', '4block', '5block']
    tallest_stable_over_time = np.zeros((logger.args.max_acquisitions, 5))
    tp = TowerPlanner(stability_mode='contains')
    for tx in range(logger.args.max_acquisitions):
        acquired_data, _ = logger.load_acquisition_data(tx)

        # For each tower, figure out when it fell over.
        for kx, k in enumerate(tower_keys):
            towers = acquired_data[k]['towers']
            for ix in range(0, towers.shape[0]):
                height = 1
                for top_id in range(1, towers.shape[1]):
                    block_tower = [Object.from_vector(towers[ix, bx, :]) for bx in range(0, top_id+1)]
                    if not tp.tower_is_constructable(block_tower):
                        break
                    height += 1
                tallest_stable_over_time[tx, height-1] += 1

    max_x = 40 + 10*logger.args.max_acquisitions
    xs = np.arange(40, max_x, 10)

    w = 10
    plt.figure(figsize=(20, 10))
    plt.bar(xs, tallest_stable_over_time[:, 0], width=w, label=1)
    for kx in range(1, 5):
        plt.bar(xs, tallest_stable_over_time[:, kx], bottom=np.sum(tallest_stable_over_time[:, :kx], axis=1), width=w, label=kx+1)
    
    plt.xlabel('Acquisition Step')
    plt.ylabel('Height of tallest stable subtower')
    plt.legend()
    plt.savefig(logger.get_figure_path('tallest_breakdown.png'))
def random_block_set(args):
    dim_range = (args.block_min_dim, args.block_max_dim)
    block_set = [
        Object.random('obj_' + str(n), dim_range=dim_range)
        for n in range(args.n_blocks)
    ]
    pkl_filename = 'block_set_' + str(args.n_blocks) + '.pkl'
    with open(pkl_filename, 'wb') as f:
        pickle.dump(block_set, f)
示例#6
0
 def get_random_pos(self, rot, block):
     """ Sample a random offset from a single corner of the platform.
     :param rot: The rotation the block will be placed with.
     :param block: Used to get the dimensions of the block to place.
     """
     # rotate the block by that rotation to get the dimensions in x,y
     new_dims = np.abs(rot.apply(block.dimensions))
     # and sample a position within the block to place on the corner of the platform
     place_pos = new_dims*(np.random.rand(3) - 0.5)
     table, _ = Object.platform()
     x, y, _ = place_pos + np.array(table.dimensions)/2
     return Position(x, y, 0)
示例#7
0
def ros_to_tower(msg):
    """ Converts a list of ROS BodyInfo messages to a tower of Object variables """
    tower = []
    for ros_blk in msg:
        # Unpack the block information
        name = ros_blk.name
        dims = (ros_blk.dimensions.x, ros_blk.dimensions.y, ros_blk.dimensions.z)
        com = (ros_blk.com.x, ros_blk.com.y, ros_blk.com.z)
        mass = ros_blk.mass
        color = (ros_blk.color.r, ros_blk.color.g, ros_blk.color.b)
        blk = Object(name, dims, mass, com, color)

        # Now create a Pose namedtuple
        pose = ros_to_pose_tuple(ros_blk.pose)
        pos = Position(pose[0][0], pose[0][1], pose[0][2])
        rot = Quaternion(pose[1][0], pose[1][1], pose[1][2], pose[1][3])
        pose_named = Pose(pos=pos, orn=Quaternion(0,0,0,1))
        blk.set_pose(pose_named)
        blk.rotation = rot
        tower.append(blk)
    return tower
def inspect_validation_set(fname):
    with open(fname, 'rb') as handle:
        val_towers = pickle.load(handle)

    tp = TowerPlanner(stability_mode='contains')
    
    towers = val_towers['5block']['towers']
    labels = val_towers['5block']['labels']

    # Check how may towers fall over at a lower block.
    for tower_vec, label in zip(towers, labels):
        if label == 0:
            tower = [Object.from_vector(tower_vec[bx, :]) for bx in range(0, tower_vec.shape[0])]
            print(tp.tower_is_constructable(tower[:1]))
示例#9
0
def sample_unlabeled_data(n_samples, block_set=None, range_n_blocks=(2, 5)):
    """ Generate n_samples random towers. For now each sample can also have
    random blocks. We should change this later so that the blocks are fixed 
    (i.e., chosen elsewhere) and we only sample the configuration.
    :param n_samples: Number of random towers to consider.
    :param block_set (optional): blocks to use in towers. generate new blocks if None
    :return: Dict containining numpy arrays of the towers sorted by size.
    """
    # initialize a dictionary of lists to store the generated data

    sampled_towers = {}
    for i in range(range_n_blocks[0], range_n_blocks[1] + 1):
        k = f'{i}block'
        sampled_towers[k] = {}
        sampled_towers[k]['towers'] = []
        sampled_towers[k]['labels'] = []
        if block_set is not None:
            sampled_towers[k]['block_ids'] = []

    min_n_blocks = range_n_blocks[0]
    max_n_blocks = min(range_n_blocks[1], len(block_set))

    # sample random towers and add them to the lists in the dictionary
    for ix in range(n_samples):
        n_blocks = np.random.randint(min_n_blocks, max_n_blocks + 1)
        # get n_blocks, either from scratch or from the block set
        if block_set is not None:
            blocks = np.random.choice(block_set, n_blocks, replace=False)
        else:
            blocks = [Object.random(f'obj_{ix}') for ix in range(n_blocks)]
        # sample a new tower
        tower = sample_random_tower(blocks)
        rotated_tower = [get_rotated_block(b) for b in tower]
        # and save that tower in the sampled_towers dict
        sampled_towers['%dblock' % n_blocks]['towers'].append(
            vectorize(rotated_tower))
        if block_set is not None:
            block_ids = [block.get_id() for block in rotated_tower]
            sampled_towers['%dblock' % n_blocks]['block_ids'].append(block_ids)

    # convert all the sampled towers to numpy arrays
    for k in sampled_towers.keys():
        sampled_towers[k]['towers'] = np.array(sampled_towers[k]['towers'])
        sampled_towers[k]['labels'] = np.zeros(
            (sampled_towers[k]['towers'].shape[0], ))
        if block_set is not None:
            sampled_towers[k]['block_ids'] = np.array(
                sampled_towers[k]['block_ids'])

    return sampled_towers
示例#10
0
def make_platform_world(p_block, action):
    """ Given a block, create a world that has a platform to push that block off of.
    :param block: The Object which to place on the platform.
    """
    platform_table, platform_leg = Object.platform()

    p_block.set_pose(Pose(ZERO_POS, Quaternion(*action.rot.as_quat())))
    block = get_rotated_block(p_block)
    block.set_pose(Pose(pos=Position(x=action.pos.x,
                                     y=action.pos.y,
                                     z=platform_table.get_pose().pos.z+platform_table.dimensions.z/2.+block.dimensions.z/2.),
                        orn=ZERO_ROT))

    return World([platform_table, block, platform_leg])
def check_validation_robustness(noise=0.001, n_attempts=10):
    """
    Try adding noise to the placement of each block in the validation set
    to see how many of the towers are robust to placement noise.
    """
    with open('learning/data/validation_towers_robust.pkl', 'rb') as handle:
    #with open('learning/data/random_blocks_(x2000)_5blocks_uniform_mass.pkl', 'rb') as handle:
        val_towers = pickle.load(handle)
    robust = {k: 0 for k in val_towers.keys()}
    tp = TowerPlanner(stability_mode='contains')
    stable_towers = copy.deepcopy(val_towers)
    unstable_towers = copy.deepcopy(val_towers)
    for k in robust.keys():
        stable_indices = []
        unstable_indices = []
        for ix in range(0, val_towers[k]['towers'].shape[0]):
            stable = True
            if val_towers[k]['labels'][ix] == 0:
                continue

            for _ in range(n_attempts):
                tower = val_towers[k]['towers'][ix, :, :].copy()
                label = val_towers[k]['labels'][ix]
                tower[:, 7:9] += np.random.randn(2*tower.shape[0]).reshape(tower.shape[0], 2)*noise

                block_tower = [Object.from_vector(tower[kx, :]) for kx in range(tower.shape[0])]

                if tp.tower_is_constructable(block_tower) != label:
                    stable = False

            if stable:
                robust[k] += 1
                stable_indices.append(ix)
            else:
                unstable_indices.append(ix)
        
        stable_towers[k]['towers'] = stable_towers[k]['towers'][stable_indices,...]
        stable_towers[k]['labels'] = stable_towers[k]['labels'][stable_indices]
        
        unstable_towers[k]['towers'] = unstable_towers[k]['towers'][unstable_indices,...]
        unstable_towers[k]['labels'] = unstable_towers[k]['labels'][unstable_indices]
        
        # with open('learning/data/stable_val.pkl', 'wb') as handle:
        #     pickle.dump(stable_towers, handle)

        # with open('learning/data/unstable_val.pkl', 'wb') as handle:
        #     pickle.dump(unstable_towers, handle)
        
        print(k, ':', robust[k], '/', val_towers[k]['towers'].shape[0] )
def check_stable_bases(logger):
    tower_keys = ['2block', '3block', '4block', '5block']
    tp = TowerPlanner(stability_mode='contains')
    for tx in range(0, 80):
        print(tx)
        towers, _ = logger.load_acquisition_data(tx)

        for k in tower_keys:
            print(k)
            for tower_vec in towers[k]['towers']:
                tower = [Object.from_vector(tower_vec[bx, :]) for bx in range(0, tower_vec.shape[0])]
                if tp.tower_is_constructable(tower[:-1]):
                    print('Stable Base')
                else:
                    print('Unstable Base')
def is_robust(orig_tower, n_attempts=10, noise=0.001):
    """ Perturb each block in the tower by 1mm multiple times and make sure the label does not change. """
    tp = TowerPlanner(stability_mode='contains')
    robust = True
    tower_vec = np.array(
        [orig_tower[bx].vectorize() for bx in range(0, len(orig_tower))])
    label = tp.tower_is_constructable(orig_tower)
    for _ in range(n_attempts):
        tower = tower_vec.copy()
        tower[:, 7:9] += np.random.randn(2 * tower.shape[0]).reshape(
            tower.shape[0], 2) * noise

        block_tower = [
            Object.from_vector(tower[kx, :]) for kx in range(tower.shape[0])
        ]

        if tp.tower_is_constructable(block_tower) != label:
            robust = False

    return robust
示例#14
0
def block_set_from_csv(args):
    block_set = []
    with open(args.csv_file) as csv_file:
        csv_reader = csv.reader(csv_file, delimiter=',')
        for row_i, row in enumerate(csv_reader):
            if row_i > 0:
                id, dimensions, com, mass = row

                block = Object(
                    'obj_' + id,
                    Dimensions(*np.multiply(string_to_list(dimensions),
                                            .01)),  # g --> kg
                    float(mass) * .001,  # g --> kg
                    Position(
                        *np.multiply(string_to_list(com), .01)),  # cm --> m
                    Color(*np.random.rand(3)))
                print(block.dimensions, block.mass, block.com)
                block_set.append(block)

    pkl_filename = args.csv_file[:-4] + '.pkl'
    with open(pkl_filename, 'wb') as f:
        pickle.dump(block_set, f)
def save_collected_tower_images(logger):
    tower_keys = ['2block', '3block', '4block', '5block']
    for tx in range(0, 20):
        towers, _ = logger.load_acquisition_data(tx)
        
        ix = np.random.randint(0, 10)
        print(towers['2block']['towers'].shape)
        start = 0 
        for k in tower_keys:
            end = start + towers[k]['towers'].shape[0]

            if ix < end:
                tower = towers[k]['towers'][ix - start, :, :]
                block_tower = [Object.from_vector(tower[bx, :]) for bx in range(tower.shape[0])]
                label = towers[k]['labels'][ix-start]
                break
            start = end

        w = World(block_tower)
        env = Environment([w], vis_sim=False, vis_frames=True)
        env.step(vis_frames=True)

        view_matrix = p.computeViewMatrixFromYawPitchRoll(distance=0.3,
                                                            yaw=45,
                                                            pitch=-10,
                                                            roll=0,
                                                            upAxisIndex=2,
                                                            cameraTargetPosition=(0., 0., 0.25))
        aspect = 100. / 190.
        nearPlane = 0.01
        farPlane = 10
        fov = 90
        projection_matrix = p.computeProjectionMatrixFOV(fov, aspect, nearPlane, farPlane)
        image_data = p.getCameraImage(200, 380, shadow=1,  viewMatrix=view_matrix, projectionMatrix=projection_matrix)
        w, h, im = image_data[:3]
        np_im = np.array(im, dtype=np.uint8).reshape(h, w, 4)[:, :, 0:3]
        #plt.imshow(np.array(np_im))
        plt.imsave(logger.get_figure_path('tower_%d_%d.png' % (tx, label)), np_im)
        env.disconnect()
示例#16
0
def active(strategy, vis=False):
    hypotheses = get_all_hypotheses()
    tp = TowerPlanner(stability_mode='contains')

    for nx in range(1, MAX_N):
        # Generate a random set of 5 blocks.
        blocks = [Object.random(f'obj_{ix}') for ix in range(NUM_BLOCKS)]

        # Choose a tower to build.
        if strategy == 'random':
            num_blocks = np.random.randint(2, NUM_BLOCKS + 1)
            tower = sample_random_tower(blocks[:num_blocks])
            tower = [get_rotated_block(b) for b in tower]
            tower = [deepcopy(b) for b in tower]
        elif strategy == 'entropy':
            tower = find_entropy_tower(blocks, hypotheses)
        else:
            raise NotImplementedError()

        # Check for consistent models.
        valid_hypotheses = []
        for h in hypotheses:
            true = tp.tower_is_stable(tower)
            pred = h(tower)
            if true == pred:
                valid_hypotheses.append(h)
        hypotheses = valid_hypotheses

        # Visualize the chosen tower and print the updated hypothesis list.
        if vis:
            TeleportAgent.simulate_tower(tower, vis=True, T=300)
            print(hypotheses)

        # Check if true model found.
        if len(hypotheses) == 1:
            break

    return nx
示例#17
0
def get_labels(samples,
               exec_mode,
               agent,
               logger,
               xy_noise,
               save_tower=False,
               label_subtowers=False):
    """ Takes as input a dictionary from the get_subset function. 
    Augment it with stability labels. 
    :param samples:
    :param exec_mode: str in ['simple-model', 'noisy-model', 'sim', 'real']
    :param agent: PandaAgent or None (if exec_mode == 'simple-model' or 'noisy-model')
    :return:
    """
    labeled_samples = {'%dblock' % k: {} for k in [2, 3, 4, 5]}
    for k in labeled_samples:
        labeled_samples[k]['towers'] = []
        labeled_samples[k]['block_ids'] = []
        labeled_samples[k]['labels'] = []

    block_placements = 0

    tp = TowerPlanner(stability_mode='contains')
    for k in samples.keys():
        n_towers, n_blocks, _ = samples[k]['towers'].shape
        labels = np.ones((n_towers, ))

        for ix in range(0, n_towers):
            print(f'Collecting tower {ix+1}/{n_towers} for {k} towers...')
            # Add noise to blocks and convert tower to Block representation.
            block_tower = []
            for jx in range(n_blocks):
                vec_block = deepcopy(samples[k]['towers'][ix, jx, :])
                if exec_mode == 'noisy-model':
                    vec_block[7:9] += np.random.randn(2) * xy_noise
                block = Object.from_vector(
                    vec_block)  # block is already rotated
                if 'block_ids' in samples[k].keys():
                    block.name = 'obj_' + str(samples[k]['block_ids'][ix, jx])
                block_tower.append(block)
            #  Use tp to check for stability.
            if exec_mode == 'simple-model' or exec_mode == 'noisy-model':

                # iterate through each subtower until it falls (is not constructable)
                subtowers = [
                    block_tower[:k_sub]
                    for k_sub in list(range(2,
                                            len(block_tower) + 1))
                ]
                for k_sub, subtower in enumerate(subtowers, 2):
                    if tp.tower_is_constructable(subtower):
                        label = 1.0
                    else:
                        label = 0.0

                    # add to labeled samples
                    labeled_samples['%dblock' % k_sub]['towers'].append(
                        samples[k]['towers'][ix, :k_sub, :])
                    if 'block_ids' in labeled_samples['%dblock' % k_sub]:
                        labeled_samples['%dblock' % k_sub]['block_ids'].append(
                            samples[k]['block_ids'][ix, :k_sub])
                    labeled_samples['%dblock' % k_sub]['labels'].append(label)

                    # save tower file
                    if save_tower:
                        if 'block_ids' in samples[k].keys():
                            logger.save_towers_data(
                                samples[k]['towers'][ix, :k_sub, :],
                                samples[k]['block_ids'][ix, :k_sub], label)
                        else:
                            logger.save_towers_data(
                                samples[k]['towers'][ix, :k_sub, :], None,
                                label)
                    # stop when tower falls
                    if label == 0.0:
                        block_placements += k_sub
                        labels[ix] = 0.0
                        break
            else:
                vis = True
                success = False
                real = (exec_mode == 'real')
                # if planning fails, reset and try again
                while not success:
                    success, label = agent.simulate_tower(block_tower,
                                                          vis,
                                                          real=real)
                    print(f"Received success: {success}, label: {label}")
                    if not success:
                        if real:
                            input(
                                'Resolve conflict causing planning to fail, then press \
                                    enter to try again.')
                            if isinstance(agent, PandaClientAgent):
                                agent.restart_services()
                        else:  # in sim
                            input(
                                'Should reset sim. Not yet handled. Exit and restart training.'
                            )
                labels[ix] = label
                if 'block_ids' in samples[k].keys():
                    logger.save_towers_data(samples[k]['towers'][ix, :, :],
                                            samples[k]['block_ids'][ix, :],
                                            labels[ix])
                else:
                    logger.save_towers_data(samples[k]['towers'][ix, :, :],
                                            None, labels[ix])
        samples[k]['labels'] = labels

    if save_tower:
        # save block placement data
        logger.save_block_placement_data(block_placements)

    if label_subtowers:
        # vectorize labeled samples and return
        for ki, k in enumerate(labeled_samples, 2):
            if labeled_samples[k]['towers'] == []:
                labeled_samples[k]['towers'] = np.zeros((0, ki, 21))
                labeled_samples[k]['block_ids'] = np.zeros((0, ki))
                labeled_samples[k]['labels'] = np.zeros(0)
            labeled_samples[k]['towers'] = np.array(
                labeled_samples[k]['towers'])
            labeled_samples[k]['block_ids'] = np.array(
                labeled_samples[k]['block_ids'])
            labeled_samples[k]['labels'] = np.array(
                labeled_samples[k]['labels'])
        return labeled_samples
    else:
        return samples
示例#18
0
文件: run.py 项目: shiyani21/stacking
    parser.add_argument('--max-height',
                        default=5,
                        type=int,
                        help='number of blocks in goal tower')

    args = parser.parse_args()

    if args.debug:
        import pdb
        pdb.set_trace()

    if args.block_set_fname is not '':
        with open(args.block_set_fname, 'rb') as f:
            block_set = pickle.load(f)
    else:
        block_set = [Object.random(f'obj_{ix}') for ix in range(args.n_blocks)]

    logger = ActiveExperimentLogger(args.exp_path)
    pre = 'discrete_' if args.discrete else ''

    ## RUN SEQUENTIAL PLANNER
    if args.method == 'sequential' or args.method == 'both':
        tower_sizes = [2, 3, 4, 5]
        tower_keys = [str(ts) + 'block' for ts in tower_sizes]
        max_height = args.max_height

        # Store regret for towers of each size.
        regrets = {k: [] for k in tower_keys}
        rewards = {k: [] for k in tower_keys}
        num_nodes = {k: [] for k in tower_keys}
        trees = []
示例#19
0
def sample_sequential_data(block_set, dataset, n_samples):
    """ Generate n_samples random towers. Each tower has the property that its
    base (the tower until the last block) is stable. To ensure this, we start
    with all the stable towers plus base cases of single block towers.
    :param block_set: List of blocks that can be used in the towers.
    :param dataset: List of current towers that have been built.
    :param n_samples: Number of random towers to consider.
    :param max_blocks
    :return: Dict containining numpy arrays of the towers sorted by size.
    """
    print('Generating data sequentially...')
    keys = ['2block', '3block', '4block', '5block']

    # initialize a dictionary of lists to store the generated data
    sampled_towers = {k: {} for k in keys}
    for k in keys:
        sampled_towers[k]['towers'] = []
        sampled_towers[k]['labels'] = []
        if block_set is not None:
            sampled_towers[k]['block_ids'] = []

    # Gather list of all stable towers. Towers should be of blocks that are rotated in "Block" format.
    stable_towers = []
    # Get all single block stable towers.
    for block in block_set:
        for orn in QUATERNIONS:
            new_block = deepcopy(block)
            new_block.pose = Pose(ZERO_POS, orn)
            rot_block = get_rotated_block(new_block)
            rot_block.pose = Pose((0., 0., rot_block.dimensions.z / 2.),
                                  (0, 0, 0, 1))
            stable_towers.append([rot_block])

    # Get all stable towers from the dataset.
    for k in keys[:3]:
        if dataset is None:
            break
        tower_tensors = unprocess(
            dataset.tower_tensors[k].cpu().numpy().copy())
        tower_labels = dataset.tower_labels[k]
        for ix, (tower_vec,
                 tower_label) in enumerate(zip(tower_tensors, tower_labels)):
            if tower_label == 1:
                block_tower = []
                for bx in range(tower_vec.shape[0]):
                    block = Object.from_vector(tower_vec[bx, :])
                    if block_set is not None:
                        block.name = 'obj_' + str(
                            int(dataset.tower_block_ids[k][ix, bx]))
                    block_tower.append(block)
                stable_towers.append(block_tower)

    # maintain block info my block name
    # TODO: this will NOT work if using random blocks
    block_lookup = {}
    for block in block_set:
        block_lookup[block.name] = block

    # Sample random towers by randomly choosing a stable base then trying to add a block.
    for ix in range(n_samples):
        # Choose a stable base.
        tower_ix = np.random.choice(np.arange(0, len(stable_towers)))
        base_tower = stable_towers[tower_ix]

        # Choose a block that's not already in the tower.
        remaining_blocks = {}
        for k in block_lookup:
            used = False
            for block in base_tower:
                if k == block.name:
                    used = True
            if not used:
                remaining_blocks[k] = block_lookup[k]

        # if we switch block sets during training then the remaining_blocks list
        # will be longer than block_set - len(base_tower)
        #assert(len(remaining_blocks) == len(block_set) - len(base_tower))

        new_block = deepcopy(np.random.choice(list(remaining_blocks.values())))

        # Choose an orientation.
        orn = QUATERNIONS[np.random.choice(np.arange(0, len(QUATERNIONS)))]
        new_block.pose = Pose(ZERO_POS, orn)
        rot_block = get_rotated_block(new_block)

        # Sample a displacement.
        base_dims = np.array(base_tower[-1].dimensions)[:2]
        new_dims = np.array(rot_block.dimensions)[:2]
        max_displacements_xy = (base_dims + new_dims) / 2.
        noise_xy = np.clip(0.5 * np.random.randn(2), -0.95, 0.95)
        rel_xy = max_displacements_xy * noise_xy

        # Calculate the new pose.
        base_pos = np.array(base_tower[-1].pose.pos)[:2]
        pos_xy = base_pos + rel_xy
        pos_z = np.sum([b.dimensions.z
                        for b in base_tower]) + rot_block.dimensions.z / 2.
        rot_block.pose = Pose((pos_xy[0], pos_xy[1], pos_z), (0, 0, 0, 1))

        # Add block to tower.
        new_tower = base_tower + [rot_block]

        if False:
            w = World(new_tower)
            env = Environment([w], vis_sim=True, vis_frames=True)
            for tx in range(240):
                env.step(vis_frames=True)
                time.sleep(1 / 240.)
            env.disconnect()
        # Save that tower in the sampled_towers dict
        n_blocks = len(new_tower)
        sampled_towers['%dblock' % n_blocks]['towers'].append(
            vectorize(new_tower))

        # save block id
        if block_set is not None:
            block_ids = [block.get_id() for block in new_tower]
            sampled_towers['%dblock' % n_blocks]['block_ids'].append(block_ids)

    # convert all the sampled towers to numpy arrays
    for k in keys:
        sampled_towers[k]['towers'] = np.array(sampled_towers[k]['towers'])
        if sampled_towers[k]['towers'].shape[0] == 0:
            sampled_towers[k]['towers'] = sampled_towers[k]['towers'].reshape(
                (0, int(k[0]), 21))
        sampled_towers[k]['labels'] = np.zeros(
            (sampled_towers[k]['towers'].shape[0], ))
        if block_set is not None:
            sampled_towers[k]['block_ids'] = np.array(
                sampled_towers[k]['block_ids'])
    return sampled_towers
示例#20
0
def sample_next_block(n_samples, bases={}, block_set=None):
    # if the dataset is empty, we are sampling the 2-block bases of the towers
    if bases == {}:
        return sample_unlabeled_data(n_samples,
                                     block_set=block_set,
                                     range_n_blocks=(2, 2))
    # if the dataset is non-empty, then for each tower in the dataset we need to sample
    # a bunch of options for the next block to be placed on top
    else:
        assert len(
            list(bases.keys())
        ) == 1, 'I want all the towers to be the same height cuz i\'m rushing'
        base_n_blocks_key = list(bases.keys())[0]
        base_n_blocks = int(base_n_blocks_key.strip('block'))
        n_towers = bases[base_n_blocks_key]['towers'].shape[0]
        n_new_blocks_per_base = np.ceil(n_samples / n_towers).astype(int)
        new_towers = []
        new_block_ids = []
        for i in range(n_towers):
            # pull out some information about the tower we're working on
            current_tower = bases[base_n_blocks_key]['towers'][i]
            top_block = current_tower[-1]
            top_block_dims = top_block[4:7]
            top_block_pos = top_block[7:10]
            top_of_tower_height = top_block_pos[2] + top_block_dims[2] / 2

            # get n_new_blocks_per_base new blocks to add on top of the tower
            if block_set is None:
                new_top_blocks = [
                    Object.random(f'obj_{ix}')
                    for ix in range(n_new_blocks_per_base)
                ]
            else:
                # we can't duplicate blocks, so only choose new top blocks from the
                # blocks that aren't already in the base
                block_ids_already_in_tower = list(
                    bases[base_n_blocks_key]['block_ids'][i])
                remaining_blocks = [
                    b for b in block_set
                    if b.name.strip('obj_') not in block_ids_already_in_tower
                ]
                new_top_blocks = sample_with_replacement(
                    remaining_blocks, k=n_new_blocks_per_base)
                # save the block ids for each of the new towers
                new_top_block_ids = [
                    b.name.strip('obj_') for b in new_top_blocks
                ]
                new_block_ids_local = np.zeros(
                    [n_new_blocks_per_base, base_n_blocks + 1])
                new_block_ids_local[:, :-1] = block_ids_already_in_tower
                new_block_ids_local[:, -1] = new_top_block_ids
                new_block_ids.append(new_block_ids_local)

            # get random rotations for each block
            orns = sample_with_replacement(QUATERNIONS,
                                           k=n_new_blocks_per_base)

            # apply the rotations to each block
            rotated_blocks = []
            for orn, block in zip(orns, new_top_blocks):
                block.pose = Pose(ZERO_POS, orn)
                rotated_blocks.append(get_rotated_block(block))

            # figure out how far each block can be moved w/ losing contact w/ the block below
            dims_xy = np.array([rb.dimensions for rb in rotated_blocks])[:, :2]
            # figure out how far each block can be moved w/ losing contact w/ the block below
            max_displacements_xy = (top_block_dims[:2] + dims_xy) / 2.
            # sample unscaled noise (clip bceause random normal can exceed -1, 1)
            noise_xy = np.clip(0.5 * np.random.randn(n_new_blocks_per_base, 2),
                               -0.95, 0.95)
            # and scale the noise by the max allowed displacement
            rel_xy = max_displacements_xy * noise_xy
            # and get the actual pos by the difference to the top block pos
            pos_xy = top_block_pos[:2] + rel_xy

            # calculate the height of each block
            pos_z = np.array([
                rb.dimensions.z / 2 + top_of_tower_height
                for rb in rotated_blocks
            ])
            pos_xyz = pos_xyz = np.hstack([pos_xy, pos_z[:, None]])

            for pos, orn, block in zip(pos_xyz, orns, new_top_blocks):
                block.pose = Pose(Position(*pos), orn)
                block.rotation = orn

            # create an array to hold all the new towers
            new_towers_local = np.zeros([
                n_new_blocks_per_base, current_tower.shape[0] + 1,
                current_tower.shape[1]
            ])
            # add the existing base of the tower to the array
            new_towers_local[:, :-1] = current_tower
            # and add the new top block to each tower
            new_towers_local[:, -1] = vectorize(new_top_blocks)
            new_towers.append(new_towers_local)

        # package the new towers into a dict of the appropriate format. include
        # block_ids if we are using a block set
        new_towers = np.concatenate(new_towers)
        new_samples = {
            'towers': new_towers,
            'labels': np.zeros(new_towers.shape[0])
        }
        if block_set is not None:
            new_samples['block_ids'] = np.concatenate(new_block_ids)

        return {f'{base_n_blocks+1}block': new_samples}
def main(vis_tower=False):
    num_towers = 500

    pw_stable = True
    # create a vector of stability labels where half are unstable and half are stable
    stability_labels = np.zeros(num_towers * 2, dtype=int)
    stability_labels[500:] = 1.

    dataset = {}
    for num_blocks in range(3, 6):
        vectorized_towers = []

        for constructable in [False, True]:
            count = 0
            while count < num_towers:
                # print the information about the tower we are about to generate
                stability_type = "con" if constructable else "uncon"
                stability_type += "/pw_stable" if pw_stable else "/pw_unstable"
                print(
                    f'{count}/{num_towers}\t{stability_type} {num_blocks}-block tower'
                )

                # generate random blocks. Use the block set if specified. otherwise
                # generate new blocks from scratch. Save the block names if using blocks
                # from the block set
                blocks = [
                    Object.random(f'obj_{ix}') for ix in range(num_blocks)
                ]
                cog_stable = np.random.choice([True, False])
                tower = build_tower(blocks,
                                    constructable=constructable,
                                    pairwise_stable=pw_stable,
                                    cog_stable=cog_stable)

                if tower is None:
                    continue
                if not is_robust(tower):
                    print('Not Robust')
                    continue

                if vis_tower:
                    w = World(tower)
                    env = Environment([w], vis_sim=True, vis_frames=True)
                    print(stability_type)
                    input()
                    for tx in range(240):
                        env.step(vis_frames=False)
                        time.sleep(1 / 240.)
                    env.disconnect()

                count += 1
                # append the tower to the list
                vectorized_towers.append(vectorize(tower))

        data = {
            'towers': np.array(vectorized_towers),
            'labels': stability_labels,
        }

        dataset[f'{num_blocks}block'] = data

    # save the generate data
    filename = 'learning/data/validation_towers_robust.pkl'
    print('Saving to', filename)
    with open(filename, 'wb') as f:
        pickle.dump(dataset, f)
示例#22
0
    def plan(self,
             blocks,
             ensemble,
             reward_fn,
             args,
             num_blocks=None,
             n_tower=None):
        #n = len(blocks)
        #max_height = 0
        #max_tower = []

        # Step (1): Build dataset of potential towers.
        tower_vectors, tower_block_ids = self.generate_candidate_towers(
            blocks, args, num_blocks, n_tower)

        # Step (2): Get predictions for each tower.
        towers = np.array(tower_vectors)
        block_ids = np.array(tower_block_ids)
        if args.planning_model == 'learned':
            # Since we are only planning for towers of a single size,
            # always use the '2block' key for simplicity. The rest currently
            # need at least some data for the code to work.
            labels = np.zeros((towers.shape[0], ))
            tower_dict = {}
            for k in self.tower_keys:
                tower_dict[k] = {}
                if k == '2block':
                    tower_dict[k]['towers'] = towers
                    tower_dict[k]['labels'] = labels
                    tower_dict[k]['block_ids'] = block_ids
                else:
                    tower_dict[k]['towers'] = towers[:5, ...]
                    tower_dict[k]['labels'] = labels[:5]
                    tower_dict[k]['block_ids'] = block_ids[:5, ...]

            tower_dataset = TowerDataset(tower_dict, augment=False)
            tower_sampler = TowerSampler(dataset=tower_dataset,
                                         batch_size=64,
                                         shuffle=False)
            tower_loader = DataLoader(dataset=tower_dataset,
                                      batch_sampler=tower_sampler)
            preds = []
            if hasattr(self.logger.args,
                       'sampler') and self.logger.args.sampler == 'sequential':
                for tensor, _ in tower_loader:
                    sub_tower_preds = []
                    for n_blocks in range(2, tensor.shape[1] + 1):
                        if torch.cuda.is_available():
                            tensor = tensor.cuda()
                        with torch.no_grad():
                            sub_tower_preds.append(
                                ensemble.forward(tensor[:, :n_blocks, :]))
                    sub_tower_preds = torch.stack(sub_tower_preds, dim=0)
                    preds.append(sub_tower_preds.prod(dim=0))
            else:
                for tensor, _ in tower_loader:
                    if torch.cuda.is_available():
                        tensor = tensor.cuda()
                    with torch.no_grad():
                        preds.append(ensemble.forward(tensor))

            p_stables = torch.cat(preds, dim=0).mean(dim=1)

        elif args.planning_model == 'noisy-model':
            n_estimate = 10
            p_stables = np.zeros(len(tower_vectors))
            for ti, tower_vec in enumerate(tower_vectors):
                # estimate prob of constructability
                results = np.ones(n_estimate)
                all_stable = 1.
                for n in range(n_estimate):
                    noisy_tower = []
                    for block_vec in tower_vec:
                        noisy_block = deepcopy(block_vec)
                        noisy_block[7:9] += np.random.randn(
                            2) * args.plan_xy_noise
                        noisy_tower.append(noisy_block)
                    block_tower = [
                        Object.from_vector(block) for block in noisy_tower
                    ]
                    if not self.tp.tower_is_constructable(block_tower):
                        results[n] = 0.
                        all_stable = 0.
                        break
                p_stables[ti] = all_stable  # np.mean(results)

        elif args.planning_model == 'simple-model':
            p_stables = np.zeros(len(tower_vectors))
            for ti, tower_vec in enumerate(tower_vectors):
                block_tower = [
                    Object.from_vector(block) for block in tower_vec
                ]
                if self.tp.tower_is_constructable(block_tower):
                    p_stables[ti] = 1.

        # Step (3): Find the tallest tower of a given height.
        max_reward, max_exp_reward, max_tower, max_stable = -100, -100, None, 0
        ground_truth = -100
        max_reward_block_ids = None
        for ix, (p, tower, tower_block_ids) in enumerate(
                zip(p_stables, towers, block_ids)):
            reward = reward_fn(tower)
            exp_reward = p * reward
            if exp_reward >= max_exp_reward:  # and p > 0.5:
                if exp_reward > max_exp_reward or (exp_reward == max_exp_reward
                                                   and p > max_stable):
                    max_tower = tower_vectors[ix]
                    max_reward = reward
                    max_exp_reward = exp_reward
                    max_stable = p
                    max_reward_block_ids = tower_block_ids

            # Check ground truth stability to find maximum reward.
            if self.tp.tower_is_constructable([Object.from_vector(tower[ix,:]) for ix in range(tower.shape[0])]) \
                and reward > ground_truth:
                ground_truth = reward

        if max_tower is None:
            print('None Found')
            max_tower = tower_vectors[0]
            max_reward = reward_fn(towers[0])

        return max_tower, max_reward, ground_truth, max_reward_block_ids
示例#23
0
def augment(all_data, K_skip, translate=False, mirror=False, vis_tower=False):
    datasets = {}
    for k_block in all_data.keys():
        num_blocks = int(k_block.strip('block'))
        #print('Augmenting %d block towers...' % num_blocks)
        data = all_data[f'{num_blocks}block']
        # load the tower data
        towers = data['towers'][::K_skip, :]
        labels = data['labels'][::K_skip]
        if 'block_ids' in data.keys() and data['block_ids'].shape != (0,):
            block_ids = data['block_ids'][::K_skip, :]
        N, K, D = towers.shape
        # calculate the number of augmented towers that will be created
        N_angles = 4
        N_shift = 4 if translate else 1
        N_mirror = 3 if mirror else 1
        tower_multiplier = N_angles * N_mirror * N_shift
        N_towers_to_add = N * tower_multiplier
        # and create new arrays to store those towers
        augmented_towers = np.zeros((N_towers_to_add, K, D))
        augmented_labels = np.zeros(N_towers_to_add)
        augmented_block_ids = np.zeros((N_towers_to_add, K))

        for ix in range(N):
            #if ix % 1000 == 0:
                #print(ix)
            original_tower = [Object.from_vector(towers[ix, jx, :]) for jx in range(num_blocks)]

            rot_towers = []
            for kx, z_rot in enumerate([0., np.pi/2., np.pi, 3*np.pi/2]):
                rot = R.from_rotvec([0., 0., z_rot])
                # rotate each block in the tower and add the new tower to the dataset
                rot_poses = rot.apply(np.array([b.pose.pos for b in original_tower]))
                rot_tower = []
                for bx in range(num_blocks):
                    rot_block = deepcopy(original_tower[bx])
                    new_pose = Pose(Position(*rot_poses[bx,:].tolist()),
                                    Quaternion(*rot.as_quat().tolist()))
                    # new_pose = Pose(Position(*rot.apply(block.pose.pos)),
                    #                 Quaternion(*rot.as_quat().tolist()))
                    rot_block.set_pose(new_pose)
                    orig_rot = R.from_quat(rot_block.rotation)
                    rot_block = get_rotated_block(rot_block)
                    rot_block.rotation = (rot*orig_rot).as_quat().tolist()
                    rot_tower.append(rot_block)
                    augmented_towers[tower_multiplier*ix + kx, bx, :] = rot_tower[bx].vectorize()            

                rot_towers.append(rot_tower)
                augmented_labels[tower_multiplier*ix + kx] = labels[ix]
                if 'block_ids' in data.keys():
                    augmented_block_ids[tower_multiplier*ix + kx, :] = block_ids[ix, :]
                # translate the base block in the tower and add after the rotated blocks
                # if translate:
                #     dx, dy = np.random.uniform(-0.2, 0.2, 2)
                #     shifted_tower = augmented_towers[(4+N_shift)*ix + kx, :].copy()
                #     # Indices 7,8 correspond to the pose.
                #     # The CoM doesn't need to be shifted because it is relative.
                #     shifted_tower[:, 7] += dx
                #     shifted_tower[:, 8] += dy

                #     augmented_towers[tower_multiplier*ix + N_angles + kx, :, :] = shifted_tower
                #     augmented_labels[tower_multiplier*ix + N_angles + kx] = labels[ix]
            
            # worlds = [World(original_tower)] + [World(tower) for tower in rot_towers] 
            # env = Environment(worlds, vis_sim=True, vis_frames=True)
            # env.step(vis_frames=True)
            # input('Next?')

            # env.disconnect()
            
            # flip the mirror the COM about the COG and negate the relative position in x
            # and y for each block. Creates a new tower that is the mirror of the original
            # tower about the x and y axes
            if mirror:
                start_index = tower_multiplier*ix
                # pull out a vector of all the towers we just added. This will be of shape
                # [N_angles x num_blocks x D]
                rot_towers = augmented_towers[start_index: start_index+N_angles, ...]

                # create a vector that will mirror a tower when we multiply it
                # indices 1 and 7 correspond to the x coordinates of COM and relative position
                # indices 2 and 8 correspond to the y coordinates of COM and relative position
                mirror_in_x = np.ones([1,1,D])
                mirror_in_y = np.ones([1,1,D])
                mirror_in_x[..., [1,7]] *= -1
                mirror_in_y[..., [2,8]] *= -1

                # add the mirrored towers to the augmented towers dataset
                augmented_towers[start_index+N_angles*1 : start_index+N_angles*2, ...] = rot_towers * mirror_in_x
                augmented_towers[start_index+N_angles*2 : start_index+N_angles*3, ...] = rot_towers * mirror_in_y
                augmented_labels[start_index:start_index+N_angles*N_mirror] = labels[ix]
                if 'block_ids' in data.keys():
                    augmented_block_ids[start_index:start_index+N_angles*N_mirror, :] = block_ids[ix, :]
            if vis_tower:
                for i in range(tower_multiplier):
                    print('VISUALIZE', ix*tower_multiplier+i, N_towers_to_add)
                    new_tower = [Object.from_vector(augmented_towers[ix*tower_multiplier+i, jx, :]) for jx in range(num_blocks)]
                    w = World(new_tower)
                    env = Environment([w], vis_sim=True, vis_frames=True)
                    for tx in range(60):
                        env.step(vis_frames=False)
                        time.sleep(1/240.)
                    env.disconnect()

        datasets[f'{num_blocks}block'] = {'towers': augmented_towers,
                                          'labels': augmented_labels}
        if 'block_ids' in data.keys():
            datasets[f'{num_blocks}block']['block_ids'] = augmented_block_ids

    return datasets
示例#24
0
"""
Test the specified action by creating a test world and performing a random action.
We also test here the the same action is applied across all particle worlds.
"""
if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--action-type', choices=['push', 'place'], required=True)
    parser.add_argument('--agent-type', choices=['teleport', 'panda'], required=True)
    args = parser.parse_args()

    # Create the block.
    true_com = Position(x=0., y=0., z=0.)
    block = Object(name='block',
                   dimensions=Dimensions(x=0.05, y=0.1, z=0.05),
                   mass=1,
                   com=true_com,
                   color=Color(r=1., g=0., b=0.))

    # Test the action for all rotations.
    for r in rotation_group():
        # Create the action.
        if args.action_type == 'push':
            action = PushAction(direction=None, 
                                timesteps=50, 
                                rot=r, 
                                block=block)
        elif args.action_type == 'place':
            action = PlaceAction(pos=None,
                                 rot=r)
        else:
def inspect_2block_towers(logger):
    """
    In the full dataset, show the distribution of features.
    """
    tower_keys = ['2block', '3block', '4block', '5block']

    # dataset = logger.load_dataset(logger.args.max_acquisitions - 1)
    # print(dataset.tower_tensors['2block'].shape)
    # plt.hist(dataset.tower_tensors['2block'][:,1,8], bins=10)
    # plt.show()

    #ensemble = logger.get_ensemble(10)
    ensemble = logger.get_ensemble(logger.args.max_acquisitions - 1)
    unlabeled = sample_unlabeled_data(10000)
    preds = get_predictions(unlabeled, ensemble)
    bald_scores = bald(preds).numpy()
    print('Best BALD')
    ixs = np.argsort(bald_scores)[::-1][:10]
    print(bald_scores[ixs])
    input()
    tp = TowerPlanner(stability_mode='contains')
    preds2 = preds[:unlabeled['2block']['towers'].shape[0], :]
    bald_scores2 = bald_scores[:unlabeled['2block']['towers'].shape[0]]
    acquire_indices = np.argsort(bald_scores2)[::-1][:50]
    # for ix in range(preds2.shape[0]):
    #     print(np.around(preds2[ix,:].numpy(), 2), np.around(bald_scores2[ix], 3))
    print('-----')
    for ix in acquire_indices:
        d = decision_distance(unlabeled['2block']['towers'][ix,:,:])
        tower = unlabeled['2block']['towers'][ix,:,:]
        l = tp.tower_is_constructable([Object.from_vector(tower[bx, :]) for bx in range(tower.shape[0])])
        print(np.around(preds2[ix,:].numpy(), 4), np.around(bald_scores2[ix], 3), d, l)

    for ix in acquire_indices:
        unlabeled['2block']['towers'][ix,1,7:8] += 0.0
    new_preds = get_predictions(unlabeled, ensemble)
    print('-----')
    for ix in acquire_indices:
        d = decision_distance(unlabeled['2block']['towers'][ix,:,:])
        print(np.around(new_preds[ix,:].numpy(), 2))
    plt.hist(unlabeled['2block']['towers'][acquire_indices,1,0])
    plt.show()
    print('-----')
    start = 0
    for k in tower_keys:
        end = start + unlabeled[k]['towers'].shape[0]
        p, b = preds[start:end, :], bald_scores[start:end]
        informative = b[b > 0.3]
        print(p.shape, informative.shape)

    accs = {k: [] for k in tower_keys}
    with open('learning/data/random_blocks_(x2000)_5blocks_uniform_mass.pkl', 'rb') as handle:
        val_towers = pickle.load(handle)

    preds = get_predictions(val_towers, ensemble).mean(1).numpy()
    dists = []
    for ix in range(0, val_towers['2block']['towers'].shape[0]):
        d = decision_distance(val_towers['2block']['towers'][ix,:,:])
        dists.append(d)
    print(len(dists))
    plt.hist(dists, bins=100)
    plt.show()

    start = 0
    for k in tower_keys:
        end = start + val_towers[k]['towers'].shape[0]
        acc = ((preds[start:end]>0.5) == val_towers[k]['labels']).mean()
        accs[k].append(acc)
        start = end
    print(accs)
def evaluate_planner(logger, blocks, reward_fn, fname, args, save_imgs=False, img_prefix=''):
    tower_keys = [str(ts)+'block' for ts in args.tower_sizes]
    tp = TowerPlanner(stability_mode='contains')
    ep = EnsemblePlanner(logger)

    # Store regret for towers of each size.
    regrets = {k: [] for k in tower_keys}
    rewards = {k: [] for k in tower_keys}

    if args.max_acquisitions is not None: 
        eval_range = range(0, args.max_acquisitions, 10)
    elif args.acquisition_step is not None: 
        eval_range = [args.acquisition_step]
    
    for tx in eval_range:
        print('Acquisition step:', tx)

        ensemble = logger.get_ensemble(tx)
        if torch.cuda.is_available():
            ensemble = ensemble.cuda()
            
        for k, size in zip(tower_keys, args.tower_sizes):
            print('Tower size', k)
            num_failures, num_pw_failures = 0, 0
            curr_regrets = []
            curr_rewards = []
            for t in range(0, args.n_towers):
                print('Tower number', t)
                
                if blocks is not None:
                    plan_blocks = np.random.choice(blocks, size, replace=False)	
                    plan_blocks = copy.deepcopy(plan_blocks)	
                else:
                    plan_blocks = [Object.random() for _ in range(size)]
                    
                tower, reward, max_reward, tower_block_ids = ep.plan(plan_blocks, 
                                                                ensemble, 
                                                                reward_fn,
                                                                args,
                                                                num_blocks=size,
                                                                n_tower=t)
                                
                block_tower = []
                for vec_block, block_id in zip(tower, tower_block_ids):
                    block = Object.from_vector(vec_block)
                    block.name = 'obj_%d' % block_id
                    block_tower.append(block)         
                                           
                # save tower info to /evaluation_towers
                if args.exec_mode is None:
                    if args.planning_model == 'noisy-model':
                        logger.save_evaluation_tower(block_tower, reward, max_reward, tx, args.planning_model, args.problem, noise=args.plan_xy_noise)
                    else:
                        logger.save_evaluation_tower(block_tower, reward, max_reward, tx, args.planning_model, args.problem)

                # perturb tower if evaluating with noisy model
                if args.exec_mode == 'noisy-model':
                    block_tower = []
                    for vec_block, block_id in zip(tower, tower_block_ids):
                        vec_block[7:9] += np.random.randn(2)*args.exec_xy_noise
                        block = Object.from_vector(vec_block)
                        block.name = 'obj_%d' % block_id
                        block_tower.append(block)     
    
                # build found tower
                if args.exec_mode == 'noisy-model' or args.exec_mode == 'simple-model':
                    if not tp.tower_is_constructable(block_tower):
                        reward = 0
                        num_failures += 1
                        if tp.tower_is_pairwise_stable(block_tower):
                            num_pw_failures += 1
                        else:
                            pairs = []
                            dists = []
                            for i in range(len(tower) - 1):
                                # check that each pair of blocks is stably individually
                                top = block_tower[i+1]
                                bottom = block_tower[i]
                                if not tp.pair_is_stable(bottom, top): 
                                    pairs.append(False)
                                else:
                                    pairs.append(True)
                                top_rel_pos = np.array(top.pose.pos) - np.array(bottom.pose.pos)
                                top_rel_com = top_rel_pos + top.com
                                dists.append((np.abs(top_rel_com)*2 - bottom.dimensions)[:2])
                            #print('Pairs:', pairs, dists)
                            
                    #print('PW Stable:', tp.tower_is_pairwise_stable(block_tower))
                    #print('Global Stable:', tp.tower_is_stable(block_tower))
                    
                    if False and reward != 0:
                        print(reward, max_reward)
                        w = World(block_tower)
                        env = Environment([w], vis_sim=True, vis_frames=True)
                        input()
                        for tx in range(240):
                            env.step(vis_frames=True)
                            time.sleep(1/240.)
                        env.disconnect()
                
                    # Note that in general max reward may not be the best possible due to sampling.
                    #ground_truth = np.sum([np.max(b.dimensions) for b in blocks])
                    #print(max_reward, ground_truth)

                    # Compare heights and calculate regret.
                    regret = (max_reward - reward)/max_reward
                    #print(reward, max_reward)
                    #print(regret)
                    curr_regrets.append(regret)
                    curr_rewards.append(reward)

            if args.exec_mode == 'noisy-model' or args.exec_mode == 'simple-model':
                regrets[k].append(curr_regrets)
                rewards[k].append(curr_rewards)

        if args.max_acquisitions is not None:
            if args.exec_mode == 'noisy-model' or args.exec_mode == 'simple-model':
                with open(logger.get_figure_path(fname+'_regrets.pkl'), 'wb') as handle:
                    pickle.dump(regrets, handle)
                    
                with open(logger.get_figure_path(fname+'_rewards.pkl'), 'wb') as handle:
                    pickle.dump(rewards, handle)
            
    # if just ran for one acquisition step, output final regret and reward
    if args.acquisition_step is not None:
        if args.exec_mode == 'noisy-model' or args.exec_mode == 'simple-model':
            final_median_regret = np.median(regrets[k][0])
            final_upper75_regret = np.quantile(regrets[k][0], 0.75)
            final_lower25_regret = np.quantile(regrets[k][0][0], 0.25)
            
            final_median_reward = np.median(rewards[k][0])
            final_upper75_reward = np.quantile(rewards[k][0], 0.75)
            final_lower25_reward = np.quantile(rewards[k][0], 0.25)
            
            final_average_regret = np.average(regrets[k][0])
            final_std_regret = np.std(regrets[k][0])
            
            final_average_reward = np.average(rewards[k][0])
            final_std_reward = np.std(rewards[k][0])
            
            print('Final Median Regret: (%f) %f (%f)' % (final_lower25_regret, final_median_regret, final_upper75_regret))
            print('Final Median Reward: (%f) %f (%f)' % (final_lower25_reward, final_median_reward, final_upper75_reward))
            
            print('Final Average Regret: %f +/- %f' % (final_average_regret, final_std_regret))
            print('Final Average Reward: %f +/- %f' % (final_average_reward, final_std_reward))
示例#27
0
def setup_panda_world(robot, blocks, xy_poses=None, use_platform=True):
    # Adjust robot position such that measurements match real robot reference frame
    robot_pose = numpy.eye(4)
    robot.set_transform(robot_pose)

    pddl_blocks = []

    full_urdf_folder = 'pb_robot/tmp_urdfs'

    if not os.path.exists(full_urdf_folder):
        os.makedirs(full_urdf_folder)

    for block in blocks:
        pb_block_fname = create_pb_robot_urdf(block, block.name + '.urdf')
        pddl_block = pb_robot.body.createBody(pb_block_fname)
        pddl_blocks.append(pddl_block)

    table_x_offset = 0.2
    floor_path = 'tamp/models/panda_table.urdf'
    shutil.copyfile(floor_path, 'pb_robot/models/panda_table.urdf')
    table_file = os.path.join('models', 'panda_table.urdf')
    pddl_table = pb_robot.body.createBody(table_file)
    pddl_table.set_point([table_x_offset, 0, 0])

    frame_path = 'tamp/models/panda_frame.urdf'
    shutil.copyfile(frame_path, 'pb_robot/models/panda_frame.urdf')
    frame_file = os.path.join('models', 'panda_frame.urdf')
    pddl_frame = pb_robot.body.createBody(frame_file)
    pddl_frame.set_point(
        [table_x_offset + 0.762 - 0.0127, 0 + 0.6096 - 0.0127, 0])

    wall_path = 'tamp/models/walls.urdf'
    shutil.copyfile(wall_path, 'pb_robot/models/walls.urdf')
    wall_file = os.path.join('models', 'walls.urdf')
    pddl_wall = pb_robot.body.createBody(wall_file)
    pddl_wall.set_point([table_x_offset + 0.762 + 0.005, 0, 0])

    # Set the initial positions randomly on table.
    if xy_poses is None:
        storage_poses = [
            (-0.4, -0.45),
            (-0.4, -0.25),  # Left Corner
            (-0.25, -0.5),
            (-0.4, 0.25),  # Back Center
            (-0.4, 0.45),
            (-0.25, 0.5),  # Right Corner
            (-0., -0.5),
            (0., -0.35),  # Left Side
            (-0., 0.5),
            (0., 0.35)
        ]  # Right Side
        print('Placing blocks in storage locations...')
        for ix, block in enumerate(pddl_blocks):
            x, y = storage_poses[ix]
            dimensions = numpy.array(block.get_dimensions()).reshape((3, 1))
            if ix < 6 and (ix not in [
                    2, 5
            ]):  # Back storage should have long side along y-axis.
                for rot in all_rotations():
                    rot_dims = numpy.abs(rot.as_matrix() @ dimensions)[:, 0]
                    if rot_dims[1] >= rot_dims[0] and rot_dims[1] >= rot_dims[
                            2]:
                        block.set_base_link_pose(((x, y, 0.), rot.as_quat()))
                        break
            else:  # Side storage should have long side along x-axis.
                for rot in all_rotations():
                    rot_dims = numpy.abs(rot.as_matrix() @ dimensions)[:, 0]
                    if rot_dims[0] >= rot_dims[1] and rot_dims[0] >= rot_dims[
                            2]:
                        block.set_base_link_pose(((x, y, 0.), rot.as_quat()))
                        break

            z = pb_robot.placements.stable_z(block, pddl_table)
            block.set_base_link_point([x, y, z])
    else:
        for i, (block, xy_pose) in enumerate(zip(pddl_blocks, xy_poses)):

            full_pose = Pose(
                Position(xy_pose.pos.x, xy_pose.pos.y, xy_pose.pos.z),
                xy_pose.orn)
            block.set_base_link_pose(full_pose)

    # Setup platform.
    if use_platform:
        platform, leg = Object.platform()
        pb_platform_fname = create_pb_robot_urdf(platform, 'platform.urdf')
        pb_leg_fname = create_pb_robot_urdf(leg, 'leg.urdf')
        pddl_platform = pb_robot.body.createBody(pb_platform_fname)
        pddl_leg = pb_robot.body.createBody(pb_leg_fname)

        rotation = pb_robot.geometry.Euler(yaw=numpy.pi / 2)
        pddl_platform.set_base_link_pose(
            pb_robot.geometry.multiply(pb_robot.geometry.Pose(euler=rotation),
                                       pddl_platform.get_base_link_pose()))
        pddl_leg.set_base_link_pose(
            pb_robot.geometry.multiply(pb_robot.geometry.Pose(euler=rotation),
                                       pddl_leg.get_base_link_pose()))

        table_z = pddl_table.get_base_link_pose()[0][2]
        pddl_leg.set_base_link_point(
            [0.7, -0.4, table_z + leg.dimensions.z / 2])
        pddl_platform.set_base_link_point([
            0.7, -0.4, table_z + leg.dimensions.z + platform.dimensions.z / 2.
        ])
    else:
        pddl_platform = None
        pddl_leg = None

    return pddl_blocks, pddl_platform, pddl_leg, pddl_table, pddl_frame, pddl_wall