This navigation sub-system was designed for autonomous guidence of the UHABS ocean recovery module. While we have not got to ocean testing, we have created a simulation for proof-of-concept.
In future iterations of this project, this code can be adapted for live use with the substitution of driver and telemetry code.
While each of the modules can be used independently, we have included a full simulation through the run_boat.py and run ground_station.py scripts. These can be configured with the provided YAML template.
The navigation path of the craft is determined by ocean current (or wind vector) maps and a classical shortest path algorithm, Dijstra's.
The currents map is given by a 2-dimensional grid of u-v component vectors along with their corresponding latitude and longitude values. This is a common format that can be found in netCDF files produced by satelite measurements or model outputs.
In order to run a shortest path algorithm on the currents map we have to transform it into a discrete network graph of nodes and verticies. This is accomplished by itterating through each grid cell, calculating its geodesic distance to its neighbors, and scaling those distances by the cell's u-v current vectors to define a directed edge weight.
The result of this is a D2Q9 lattice, on which we can run our classical shortest path algorithm.
Because we have been unable to run ocean testing. We simulated runs using the same technique as above. For each timestep the ocean craft takes an azimuth based on its shortest path (with a set rate of propulsion), its position at the next timestep is determined by summing that propulsion vector with the u-v component vectors of the currents it passes over. At the begining of the next timestep the shortest path is recalculated and another azimuth is taken.
This is a similar technique to that used in the lattice boltzmann method—though it does not need to account for collisions—and likewise, its implementation can be highly-parallelized.