Пример #1
0
    def invalid_region(self, position):
        """
            Invalidates the region surrounding a single block.
            This should be used after the robot changes the environment for an reason.
        """
        self.invalid_cache()
        invalidation_size = INVALIDATION_REGION / 2

        for cluster in self.predict_block_locations():
            if util.get_distance(cluster.coord, position) < cluster.radius:
                invalidation_size = max(invalidation_size, cluster.radius)
                if(cluster.known_index is not None):
                    print("[info] Block in invalidation cluster, removing")
                    self._confirmed_blocks.pop(cluster.known_index)

        # Also delete overlapping blocks if they are not in a cluster
        for index in range(len(self._confirmed_blocks) - 1, 0, -1):
            block_pos, _ = self._confirmed_blocks[index]
            if util.get_distance(block_pos, position) < invalidation_size:
                # Purge this block
                print("[info] Block in invalidation region, removing")
                self._confirmed_blocks.pop(index)

        invalid_min = _to_screenspace(np.array(position) - invalidation_size)
        invalid_max = _to_screenspace(np.array(position) + invalidation_size)
        invalid_size = tuple(invalid_max - invalid_min)

        self._probability_sum[invalid_min[0]:invalid_max[0], invalid_min[1]:invalid_max[1]] = np.zeros(invalid_size)
        self._probability_count[invalid_min[0]:invalid_max[0], invalid_min[1]:invalid_max[1]] = np.ones(invalid_size)
Пример #2
0
    def invalid_region(self, block_locations, position, invalidation_size):
        """
            Invalidates the region surrounding a single block.
            This should be used after the robot changes the environment for an reason.
        """

        # Remove each block in a known position
        for cluster in block_locations:
            invalidation_radius = max(invalidation_size, cluster.radius)

            if util.get_distance(cluster.coord, position) < invalidation_radius:
                if(cluster.known_block is not None):
                    if cluster.known_block not in self.confirmed_blocks:
                        util.log_msg(util.Logging.WARNING, "unexpected block not in confirmed_blocks")
                        continue

                    util.log_msg(util.Logging.WARNING, "Block in invalidation cluster, removing")
                    self.confirmed_blocks.remove(cluster.known_block)

        # Also delete overlapping blocks if they are not in a cluster
        for index in range(len(self.confirmed_blocks) - 1, 0, -1):
            block_pos, _ = self.confirmed_blocks[index]
            if util.get_distance(block_pos, position) < invalidation_size:
                # Purge this block
                util.log_msg(util.Logging.INFO, "Block in invalidation region, removing")
                self.confirmed_blocks.pop(index)
Пример #3
0
    def predict_block_locations(self, cluster_candidates, occupancy_map):

        # Smooth clusters into single block
        clusters = get_cluster_average(cluster_candidates, occupancy_map)

        # Some block colors are known, mark these in the clusters object
        for block in self.confirmed_blocks:
            block_location, block_color = block
            # Check if block is in radius of any clusters
            color_consumed = False

            # NOTE: this is a back hackfix
            ClosestCluster = namedtuple("ClosestBlock", ["distance", "cluster"])
            closest_cluster = ClosestCluster(np.inf, None)

            for cluster in clusters:
                cluster_distance = util.get_distance(cluster.coord, block_location)

                if cluster_distance < mapping.CLUSTER_PROXIMITY_THRESHOLD:
                    if (cluster.color is not None):
                        util.log_msg(util.Logging.WARNING, "Multiple blocks identified in the same cluster")

                    color_consumed = True
                    cluster.assign_known_color(block)

                if cluster_distance < closest_cluster.distance:
                    closest_cluster = ClosestCluster(
                        cluster_distance,
                        cluster
                    )

            if not color_consumed:
                # Hackfix, no cluster identified properly, assign to closest one

                if closest_cluster.distance < mapping.CLUSTER_FUDGE_DISTANCE:
                    closest_cluster.cluster.assign_known_color(block)
                    continue

                new_cluster = ClusterLocation(
                    block_location, 20, 10, mapping.BLOCK_OBSTACLE_WIDTH
                )
                new_cluster.assign_known_color(block)
                clusters.append(new_cluster)

        # Ensure no clusters are inside the spawn region
        filtered_clusters = []
        for cluster in clusters:
            # Each robot spawn position is defined here
            for robot_name in util.ROBOT_SPAWN:
                spawn_pos = util.ROBOT_SPAWN[robot_name]
                if util.get_distance(spawn_pos, cluster.coord) < mapping.SPAWN_REGION_RADIUS:
                    break  # Block is inside spawn region, don't recollect it
            else:
                # Block was not inside any spawn radius, lets save it
                filtered_clusters.append(cluster)

        return filtered_clusters
Пример #4
0
    def vote_for_route(self, new_waypoints):
        # Check if routes are identical. This must be right, stick with it
        if new_waypoints == self.waypoints:
            self.votes = min(100, self.votes + 5)
            return

        routes_different = False
        if len(new_waypoints) < len(self.waypoints):
            # Check if the route just got shorter (or its completely different)
            for new_p, old_p in zip(new_waypoints[::-1], self.waypoints[::-1]):
                if GOAL_TOLERANCE > util.get_distance(new_p, old_p):
                    routes_different = True
                    break

            if not routes_different:
                # Route is just the same (but shorter)
                self.waypoints = new_waypoints
                return

        # Routes are certainly different, vote against current route
        if len(new_waypoints) < len(self.waypoints):
            self.votes -= 5  # Shorter routes are preferable
        else:
            self.votes -= 2

        # If the current route is unpopular, change it
        if self.votes < 0:
            self.waypoints = new_waypoints
            self.votes = 50
Пример #5
0
    def navigate_waypoints(self,
                           positioning_system,
                           tolerance=0.03,
                           reverse=False):
        """
            For this tick, attempt to navigate towards the next waypoint.
            Returns True if the final waypoint has been reached.
        """
        if self.__waypoint_index >= len(self.__waypoints):
            self.__waypoints = []
            self.__waypoint_index = 0
            return True

        target_waypoint = self.__waypoints[self.__waypoint_index]
        current_position = positioning_system.get_2D_position()

        self.navigate_toward_point(positioning_system, target_waypoint,
                                   reverse)

        if util.get_distance(current_position, target_waypoint) < tolerance:
            self.__waypoint_index += 1

        if self.__waypoint_index == len(self.__waypoints):
            self.__waypoints = []
            self.__waypoint_index = 0
            return True
        return False
Пример #6
0
    def predict_block_locations(self,
                                robot_states,
                                generate_fake_blocks=False):
        """
            Returns a list of ClusterLocation objects, containing the positions of blocks
            and the relative certainties.
        """
        if self.__cache_block_locations is not None:
            return self.__cache_block_locations

        occupancy_map = self.get_occupancy_map()
        cluster_candidates = self.fuzzy_mapping.generate_potential_clusters()

        block_locations = self.hard_mapping.predict_block_locations(
            cluster_candidates, occupancy_map)

        if generate_fake_blocks:
            # Try to avoid stalemates by patrolling
            for block in PATROL_COORDINATES:
                for robot_name in robot_states:
                    # Robots are near, phantom block is now a problem
                    if util.get_distance(robot_states[robot_name].position,
                                         block) < PATROL_DECAY:
                        break
                else:
                    block_locations.append(ClusterLocation(block, 1, 1, 0.4))

        self.__cache_block_locations = block_locations
        return self.__cache_block_locations
Пример #7
0
    def request_block_dropoff(self, robot_name):
        """
            The robot is holding a block, ensure it returns to its spawn position.
            If the robot is close enough for mapping to be useless, let the robot takeover.
        """

        robot_position = self.robot_states[robot_name].position

        # Robot is close enough to handle everything by itself
        destination_distance = util.get_distance(robot_position, util.ROBOT_SPAWN[robot_name])
        if destination_distance < ROBOT_TAKEOVER_DISTANCE:
            self.robot_paths[robot_name].reset_votes()
            self.radio.send_message(protocol.AskRobotDeposit(
                robot_name, util.ROBOT_SPAWN[robot_name]
            ))
            return

        # More complex routes needs to avoid collisions
        closest_path = self.pathfinding.get_spawn_path(
            robot_name, self.robot_states, self.robot_paths, robot_position
        )

        # Check if it was possible to find a path
        if not closest_path.success:
            # Send last waypoint message anyway
            # But ensure at least part of it is legal
            self.robot_paths[robot_name].waypoints = self.pathfinding.trim_legal_waypoints(
                self.robot_paths[robot_name].waypoints,
                robot_name,
                self.robot_states,
                self.robot_paths,
                True
            )

            if len(self.robot_paths[robot_name].waypoints) > 0:
                self.radio.send_message(protocol.GiveRobotTarget(
                    robot_name, self.robot_paths[robot_name].waypoints[0]
                ))
            return

        # Begin navigating back home
        # Navigate to most popular waypoint
        self.robot_paths[robot_name].vote_for_route(closest_path.waypoints)

        if not closest_path.require_popout:
            self.robot_paths[robot_name].crop_waypoints(self.robot_states[robot_name].position)

        if len(self.robot_paths[robot_name].waypoints) > 0:
            self.radio.send_message(protocol.GiveRobotTarget(
                robot_name, self.robot_paths[robot_name].waypoints[0]
            ))
        else:
            self.robot_paths[robot_name].reset_votes()
Пример #8
0
    def crop_waypoints(self, robot_position):
        """
            Dynamically updates the robot's waypoints preventing circling.
            NOTE: The robot is allowed to 'skip' waypoints
        """
        current_waypoint = -1
        for index, waypoint in enumerate(self.waypoints):
            if util.get_distance(waypoint, robot_position) < WAYPOINT_TOLERANCE:
                # Robot is close enough to waypoint to skip straight to this point in its path
                current_waypoint = index + 1

        if current_waypoint != -1:
            # New waypoint was found, skip to this point
            self.waypoints = self.waypoints[current_waypoint:]
Пример #9
0
    def tick(self):
        # Check for robot messages, take an necessary actions
        self.process_robot_messages()
        self.mapping_controller.output_to_displays()

        # If robots are navigating to a block, ensure it still exists
        block_locations = self.mapping_controller.predict_block_locations()
        for robot_name in self.robot_targets:
            block_still_exists = False
            target_position = self.robot_targets[robot_name]
            if not target_position.is_block:
                continue  # Only check if blocks exist for now

            # Loop through all known blocks
            for block in block_locations:
                if util.get_distance(
                        block.coord,
                        target_position.coord) < APPROXIMATE_POSITION:
                    block_still_exists = True

            if not block_still_exists and \
                    util.get_distance(self.robot_positions[robot_name], target_position.coord) > NO_TURNING_BACK_REGION:
                print("[info] purging block location")
                self.radio.send_message(protocol.RemoveWaypoints(robot_name))
Пример #10
0
    def __call__(self):
        if self.opening_pincer:
            return self.pincer_controller.open_pincer()

        current_position = self.positioning_system.get_2D_position()

        if util.get_distance(current_position,
                             self.dropoff_target) < DROPOFF_RADIUS:
            is_aligned = self.drive_controller.turn_toward_point(
                self.positioning_system, self.dropoff_target,
                DROPOFF_ANGLE_TOLERANCE)
            if is_aligned:
                self.drive_controller.halt()
                self.opening_pincer = True

        else:
            self.drive_controller.navigate_toward_point(
                self.positioning_system, self.dropoff_target)

        return False
Пример #11
0
    def request_block_collection(self, robot_name):
        """
            Find the closest block of the correct color, ask the robot to drive towards it.
            If the robot is close enough for mapping to be useless, let the robot takeover.
        """
        robot_position = self.robot_states[robot_name].position

        block_locations = self.mapping_controller.predict_block_locations(
            self.robot_states, self.should_patrol(robot_name))
        closest_path = self.pathfinding.get_nearest_block_path(
            block_locations, robot_name, self.robot_states, self.robot_paths,
            robot_position
        )

        # Check if map was detailed enough to find path
        if not closest_path.success:
            # Send last waypoint message anyway

            self.robot_paths[robot_name].waypoints = self.pathfinding.trim_legal_waypoints(
                self.robot_paths[robot_name].waypoints,
                robot_name,
                self.robot_states,
                self.robot_paths,
                False
            )
            if len(self.robot_paths[robot_name].waypoints) > 0:
                self.radio.send_message(protocol.GiveRobotTarget(
                    robot_name, self.robot_paths[robot_name].waypoints[0]
                ))
            return

        # Should to robot switch over to this new path?
        self.robot_paths[robot_name].vote_for_route(closest_path.waypoints)
        if closest_path.require_popout:
            self.robot_paths[robot_name].crop_waypoints(self.robot_states[robot_name].position)

        chosen_path = self.robot_paths[robot_name].waypoints
        if len(chosen_path) == 0:
            self.robot_paths[robot_name].reset_votes()
            return

        next_waypoint = chosen_path[0]
        last_waypoint = chosen_path[-1]

        # Should the robot fine tune this part itself?
        target_distance = util.get_distance(robot_position, last_waypoint)
        if target_distance < ROBOT_TAKEOVER_DISTANCE:
            self.radio.send_message(protocol.AskRobotSearch(
                robot_name, last_waypoint
            ))

            return

        # Robot needs better instructions, it cannot do this by itself

        # Sometimes waypoints will flicker and robot will be trapped
        # Smooth neighboring waypoints to minimize this
        if len(chosen_path) > 1:
            next_waypoint = (
                0.5 * (next_waypoint[0] + chosen_path[1][0]),
                0.5 * (next_waypoint[1] + chosen_path[1][1])
            )

        # Send the robot its new target position (TODO prevent collisions)
        self.radio.send_message(protocol.GiveRobotTarget(
            robot_name, next_waypoint
        ))
Пример #12
0
def get_cluster_average(cluster_candidates, weights):
    """
        Groups together True pixels and returns a list of ClusterLocation objects.
        Output is entirely in world coordinates.
    """

    # Err, comment to explain this absolute mess of a line
    intense_coords = dict(zip(zip(*np.where(cluster_candidates)), itertools.repeat(False)))
    strides = np.array([(1, 0), (-1, 0), (0, 1), (0, -1), (1, 1), (-1, -1), (1, -1), (-1, 1)])

    # Almost functional iterator
    def find_cluster_coords(coord):
        # Needs to be a tuple for indexing
        index = tuple(coord)
        if not util.tuple_in_bound(index, mapping.MAP_RESULTION) or \
                not cluster_candidates[index] or intense_coords[index]:
            # Block has already been counted or shouldn't be included
            return []

        # Mark this pixel to avoid repeated counting
        intense_coords[index] = True
        return list(reduce(
            lambda old, stride: np.concatenate((old, find_cluster_coords(coord + stride))),
            strides, coord
        ))

    # Now finally, group the clusters
    clusters_coords = []
    for coord in intense_coords:
        if intense_coords[coord] or len(coord) == 0:
            continue  # Already processed
        cluster_coords = np.reshape(find_cluster_coords(np.array(coord)), (-1, 2)).astype(int)
        clusters_coords.append(cluster_coords)

    # Now do some statistics to find centers
    cluster_locations = []  # list of BlockLocation
    for cluster_coords in clusters_coords:
        if len(cluster_coords) < mapping.CLUSTER_THRESHOLD:
            continue

        total_weight = 0
        cluster_sum = np.array((0.0, 0.0))
        for coord in cluster_coords:
            total_weight += weights[tuple(coord)]
            cluster_sum += coord * weights[tuple(coord)]

        # Create a cluster representing the average properties of these coordinates
        # Convert everything to worldspace
        average_position = mapping._to_worldspace(cluster_sum / total_weight)

        radius = 0
        for coord in cluster_coords:
            world_coord = mapping._to_worldspace(coord)
            radius = max(radius, util.get_distance(average_position, world_coord))

        cluster_locations.append(ClusterLocation(average_position, len(cluster_coords), total_weight, radius))

    # Merge nearby clusters
    merged_clusters = []
    for cluster in cluster_locations:
        for large_cluster in merged_clusters:
            cluster_distance = util.get_distance(large_cluster.coord, cluster.coord)

            # Test if clusters are close
            if cluster_distance < mapping.CLUSTER_OVERLAP_THRESHOLD * (large_cluster.radius + cluster.radius):
                # Clusters overlap, take a weighted average and change the original cluster
                large_cluster.merge_cluster(cluster)
                break
        else:
            # Cluster could not be merged into another cluster, add it to merged_clusters
            merged_clusters.append(cluster)

    return merged_clusters
Пример #13
0
    def calculate_route(self, robot_name, arena_map, start, goal):
        """
            Uses A* algorithm and knowledge of current path of other robot, along with obstacles
            in arena to avoid collisions and find shortest path and convert that into a list of
            waypoints.
        """
        walks = ((1, 0, 1), (1, 1, 2**0.5), (0, 1, 1), (-1, 1, 2**0.5),
                 (-1, 0, 1), (-1, -1, 2**0.5), (0, -1, 1), (1, -1, 2**0.5))
        distances = -np.ones((arena_map.shape), dtype=np.float64)
        directions = -np.ones((arena_map.shape), dtype=np.int32)
        path_mask = np.zeros((arena_map.shape), dtype=np.bool)
        death_mask = np.zeros((arena_map.shape), dtype=np.bool)
        distances[start] = 0
        # PriorityQueue to order squares to visit by their distance + manhatte_distance to the goal
        # meaning that it can find the optimal route without exploring the whole map
        # Format: (distance from start + manhattan distance to goal, (x, z))
        to_explore = PriorityQueue()
        to_explore.put((0, start))

        # create a bool map, where an obstacle is True and free space is False
        arena_map = np.invert(arena_map)

        # add the locations of already delivered blocks to the "Do Not Travel" areas
        death_mask = death_mask | dilate(self.mask_delivered_blocks(arena_map),
                                         7)

        # leave a safety margin around areas of "Do Not Travel"
        arena_map = arena_map | dilate(arena_map, 8)

        # remove the area around the current target location, to eliminate the block to pick
        # up from the mask of areas not to be traversed
        arena_map[max(0, goal[0] - 10):min(arena_map.shape[0] - 1, goal[0] +
                                           10),
                  max(0, goal[1] - 10):min(arena_map.shape[1] - 1, goal[1] +
                                           10)] = False

        # add the paths of the other robot to the "Do Not Travel" areas
        for other_name in self.path_masks:
            # Dilate all other robot paths
            if other_name == robot_name:
                continue
            death_mask = death_mask | dilate(self.path_masks[other_name], 8)

        self.send_to_display(arena_map)
        # explore the map with added heuristic until it is all explored or you have reached the goal
        while not to_explore.empty():
            _, (cur_x, cur_z) = to_explore.get()
            if (cur_x, cur_z) == goal:
                break
            for direction, (x, z, dist) in enumerate(walks):
                # check if movement is in map
                if not (0 <= cur_x + x < arena_map.shape[0]
                        and 0 <= cur_z + z < arena_map.shape[1]):
                    continue
                if death_mask[cur_x + x, cur_z + z]:
                    continue
                if arena_map[cur_x + x, cur_z +
                             z] == 1:  # check if movement is in obstacle
                    dist *= 1000

                # if already processed and too expensive, skip reevaluation
                if distances[cur_x + x, cur_z + z] != -1:
                    if distances[cur_x, cur_z] + dist >= distances[cur_x + x,
                                                                   cur_z + z]:
                        continue

                directions[cur_x + x, cur_z + z] = direction
                manhattan_dist = util.get_distance((cur_x + x, cur_z + z),
                                                   goal)
                distances[cur_x + x,
                          cur_z + z] = distances[cur_x, cur_z] + dist
                # store the newly reached square ready to explore its neighbours
                to_explore.put(
                    (distances[cur_x + x, cur_z + z] + manhattan_dist,
                     (cur_x + x, cur_z + z)))
        else:
            return float('inf'), [], None, path_mask.copy()  # No path found

        # Create a list of waypoints for the robot to navigate along
        waypoints = []

        # Store location where robot needs to stop in order to pick up or drop off
        # block at desired location. Required due to robot position being a few cm
        # back compared to block position
        release_pos = None

        cur_x, cur_z = goal
        path_mask[goal] = 1
        direction = directions[goal]

        # backtrack from the goal following the saved directions used to reach each square
        while (cur_x, cur_z) != start:
            cur_x -= walks[direction][0]
            cur_z -= walks[direction][1]

            goal_dist = distances[goal] - distances[cur_x, cur_z]
            if 3 < goal_dist < 5:
                release_pos = self.grid_to_world_coord((cur_x, cur_z))
            if 9 < goal_dist < 11:
                waypoints = [self.grid_to_world_coord((cur_x, cur_z))]
            path_mask[cur_x, cur_z] = 1
            if directions[cur_x, cur_z] != direction:
                waypoints.append(self.grid_to_world_coord((cur_x, cur_z)))
                direction = directions[cur_x, cur_z]

        return distances[goal], waypoints, release_pos, path_mask.copy()
Пример #14
0
    def predict_block_locations(self):
        if self.__cache_block_locations is not None:
            return self.__cache_block_locations

        GRID_SIZE = (
            int(MAP_RESULTION[0] / 5),
            int(MAP_RESULTION[1] / 5)
        )

        # Identify all clusters
        intensity_map = np.empty(MAP_RESULTION)
        occupancy_map = self.get_occupancy_map()
        mean_amplitude = np.quantile(occupancy_map, 0.99)

        # Account for local variation in intensity
        for x in range(0, MAP_RESULTION[0], 16):
            for y in range(0, MAP_RESULTION[1], 16):
                map_segment = occupancy_map[x:x + GRID_SIZE[0], y:y + GRID_SIZE[1]]
                peak_amplitude = np.quantile(map_segment, 0.98)
                intensity_map[x:x + GRID_SIZE[0], y:y + GRID_SIZE[1]] = map_segment > peak_amplitude

        cluster_candidates = np.logical_and(
            intensity_map,
            np.logical_and(
                occupancy_map > mean_amplitude,
                occupancy_map > FORCE_INVAILD_THRESHOLD
            )
        )

        # Smooth clusters into single block
        clusters = get_cluster_average(cluster_candidates, occupancy_map)

        # Some block colors are known, mark these in the clusters object
        for index, (block_location, block_color) in enumerate(self._confirmed_blocks):
            # Check if block is in radius of any clusters
            color_consumed = False

            # NOTE: this is a back hackfix
            ClosestCluster = namedtuple("ClosestBlock", ["distance", "cluster"])
            closest_cluster = ClosestCluster(np.inf, None)

            for cluster in clusters:
                cluster_distance = util.get_distance(cluster.coord, block_location)

                if cluster_distance < CLUSTER_PROXIMITY_THRESHOLD:
                    if (cluster.color is not None):
                        print("[Warning] Multiple blocks identified in the same cluster")

                    if color_consumed:
                        print("[Warning] The same blocks has been assigned to multiple clusters")

                    color_consumed = True
                    cluster.assign_known_color(index, block_color)

                if cluster_distance < closest_cluster.distance:
                    closest_cluster = ClosestCluster(
                        cluster_distance,
                        cluster
                    )

            if not color_consumed:
                # Hackfix, no cluster identified properly, assign to closest one
                # print("[Warning] Block color has been identified but does not exist on map")
                # closest_cluster.cluster.assign_known_color(index, block_color)
                new_cluster = ClusterLocation(
                    block_location, 20, 10, BLOCK_OBSTACLE_WIDTH
                )
                new_cluster.assign_known_color(index, block_color)
                clusters.append(new_cluster)

        self.__cache_block_locations = clusters
        return self.__cache_block_locations
Пример #15
0
    def __calculate_route(self, robot_name, robot_states, other_paths,
                          start_pos, goal_pos, is_returning):
        """
            Uses A* algorithm and knowledge of current path of other robot, along with obstacles
            in arena to avoid collisions and find shortest path and convert that into a list of
            waypoints.
        """
        start, goal = tuple(_to_screenspace(start_pos)), tuple(
            _to_screenspace(goal_pos))

        # Setup the pathfinding
        UNIT_STEP = 2
        DIAGONAL_STEP = 3  # 3/2 ≈ √2
        walks = np.array([(1, 0, UNIT_STEP), (1, 1, DIAGONAL_STEP),
                          (0, 1, UNIT_STEP), (-1, 1, DIAGONAL_STEP),
                          (-1, 0, UNIT_STEP), (-1, -1, DIAGONAL_STEP),
                          (0, -1, UNIT_STEP), (1, -1, DIAGONAL_STEP)],
                         dtype=np.int32)

        pathfinding_map = self._simple_map.copy()
        pathfinding_map[goal[0] - 1:goal[0] + 2,
                        goal[1] - 1:goal[1] + 2] = True
        other_robot_map = self._generate_cooperation_map(
            robot_name, robot_states, other_paths, is_returning)

        popout_position = None
        # Ensure the starting point is correct
        if not pathfinding_map[start] or not other_robot_map[start]:
            # The robot is in an illegal square, it must leave this square before continuing.
            popout_position = self.get_popout_position(
                robot_name, start_pos, pathfinding_map & other_robot_map)
            if popout_position is not None:
                start = tuple(_to_screenspace(popout_position))

        if robot_name == "Small":
            pixels = util.get_intensity_map_pixels(pathfinding_map
                                                   & other_robot_map)
            pixels[start] = (255, 0, 0)
            pixels[goal] = (0, 0, 255)
            util.display_numpy_pixels(self._display_navigation, pixels)

        costs = np.inf * np.ones(SIMPLIFIED_RESOLUTION)
        directions = np.empty(SIMPLIFIED_RESOLUTION, dtype=np.int32)

        # PriorityQueue to order squares to visit by their distance + manhatte_distance to the goal
        # meaning that it can find the optimal route without exploring the whole map
        # Format: (distance from start + manhattan distance to goal, (x, z))
        to_explore = [(0, start)]
        costs[start] = 0

        # explore the map with added heuristic until it is all explored or you have reached the goal
        while len(to_explore) > 0:
            _, current_pos = heapq.heappop(to_explore)

            for direction_id, (x, z, step_dist) in enumerate(walks):
                new_pos = (current_pos[0] + x, current_pos[1] + z)
                # check if movement is in map
                if not util.tuple_in_bound(new_pos, SIMPLIFIED_RESOLUTION):
                    continue

                # if already processed and too expensive, skip reevaluation
                new_cost = costs[current_pos] + step_dist

                if not other_robot_map[new_pos]:
                    # Avoid all robot/ robot collisions
                    continue

                # check if movement is in obstacle
                # Targets cannot be obstacles
                if new_pos != goal and (not pathfinding_map[new_pos]):
                    new_cost += 30
                    continue

                if new_cost >= costs[new_pos] or new_cost >= costs[goal]:
                    continue

                manhattan_dist = abs(new_pos[0] - goal[0]) + abs(new_pos[1] -
                                                                 goal[1])
                costs[new_pos] = new_cost
                directions[new_pos] = direction_id
                # store the newly reached square ready to explore its neighbours
                heapq.heappush(to_explore,
                               (new_cost + manhattan_dist, new_pos))

        # Check if pathfinding succeeded
        if np.isinf(costs[goal]):
            # if popout_position is not None:
            #    return PathfindingResult([popout_position], 1000, goal_pos, True)

            return PathfindingResult.NotFound(goal_pos)

        # Create a list of waypoints for the robot to navigate along
        waypoints = [tuple(goal_pos)]

        # backtrack from the goal following the saved directions used to reach each square
        current_coord = goal
        direction_index = directions[current_coord]
        while current_coord != start:
            current_coord = (current_coord[0] - walks[direction_index][0],
                             current_coord[1] - walks[direction_index][1])

            # Bad edge case here, the algorithm really should have accommodated this from the start
            if current_coord == start:
                break

            # Output a minimum number of waypoint to allow better control
            waypoints.append(tuple(_to_worldspace(current_coord)))
            direction_index = directions[current_coord]

        # Get the pathfinding error (to ensure true closest block is selected)
        if goal != start:
            penultimate_coord = (goal[0] - walks[directions[goal]][0],
                                 goal[1] - walks[directions[goal]][1])
            penultimate_cost = costs[penultimate_coord]
            cost_approximation = penultimate_cost + util.get_distance(
                goal_pos, _to_worldspace(penultimate_coord))

        else:
            cost_approximation = costs[goal]

        if popout_position is None:
            return PathfindingResult(waypoints[::-1], cost_approximation,
                                     goal_pos, False)

        return PathfindingResult(waypoints[::-1], cost_approximation, goal_pos,
                                 False)