When extracting obstacle coordinates from a lidar point cloud map in a D* algorithm implementation, the initial approach yields nested tuples with empty arrays for rows containing no obstacles. The code snippet below illustrates this method:

obstacle_points = []
for row_index in range(map_length):
    obstacle_points.append((row_index, np.where(map_grid[row_index] == 255)))

This produces output where each tuple contains a row index and a nested array of column indices where obstacles are detected, with empty arrays for obstacle-free rows. For example:

[(0, (array([], dtype=int64),)), (1, (array([], dtype=int64),)), ..., (18, (array([42, 43, 44, 45, 46, 47, 48, 49, 50]),))]

To restructure this into a flat list of coordinate pairs suitable for further processing, the data needs to be flattened. Each row index must be paired individually with every corresponding column index from its array, excludnig empty entries. The transformation can be achieved with the following restructured logic:

flattened_coordinates = []
for row_data in obstacle_points:
    current_row = row_data[0]
    column_indices = row_data[1][0]
    for col in column_indices:
        flattened_coordinates.append((current_row, col))

This results in a clean list of tuples, such as [(18, 42), (18, 43), ..., (22, 50)], where each tuple represents a single obstacle coordinate, eliminating nested structures and empty values.