Tutorials

To be able to plan, you will need:

• A map of the environment in which the robot will operate. A map created using ros2 slam_toolbox is recommended.

• A topological graph of the environment. See below for instructions on how to create an occupancy map.

• A CLiFF map learned from human motion data collected in the environment. See <https://github.com/tkucner/CLiFF-map-matlab> for instructions on how to create a CLiFF map.

• A dataset of human motion data collected in the environment.

Creating an Occupancy Map

To create an occupancy map, once you have a map of the environment you will need:

• To instantiate a predictor object with the map file and CLiFF map file as inputs. (see CliffPredictor class in src/congestion_coverage_plan/cliff_predictor/CliffPredictor.py)

• To instantiate an OccupancyMap object with the predictor object as input. (see OccupancyMap class in src/congestion_coverage_plan/occupancy_map/OccupancyMap.py)

Once instantiated, you can use the various functions of the OccupancyMap to add vertices and edges to the topological graph. An example of adding vertices and edges is shown below:

occupancy_map.add_vertex(vertex_id, posx, posy)
occupancy_map.add_edge_with_id(edge_id, start_vertex_id, end_vertex_id)

After adding all vertices and edges, you can set edge limits using the function add_edge_limit(edge_id, level) where level is a dictionary of congestion levels to be used as limits for the edge with id edge_id. An example of setting edge limits is shown below:

edges = occupancy_map.get_edges()
level = {
    'low': [0,1],
    'medium': [1,3],
    'high': [3,500]
}
for edge_key in edges:
    occupancy_map.add_edge_limit(edges[edge_key].get_id(), level)

After the creation of the topological map, you can compute the average edge traverse time using the function compute_average_edge_traverse_times().

Finally, you can save the occupancy map to a file using the function save_occupancy_map(file_path).

Running the algorithm

Once you have created and saved the occupancy map, you can run the congestion-aware planning algorithm using the occupancy map file, CLiFF map file, and human motion dataset as inputs.

To instantiate the planner, you will need to instantiate a solver (in this case the LRTDP solver) with the occupancy map (which takes also the cliff predictor as input) and the initial state defined as an MDP state.