Computing and visualizing robust plans

This project aims to provide a functional framework for implementation, evaluation and visualization of strategies used to generate robust plans to move an agent in a non-deterministic environment.

Command line interface

• -g, --graph: path to text file with graph
• -d, --direction: direction of disturbances in the graph (e.g. ublr - up, bottom, left, right)
• -f, --force: force interval of disturbances (e.g. 1-3 - disturbances with force 1, 2 and 3, default 1-1)
• -n, --count: count of random origin/destination pairs to test (default 5)
• -l, --length: minimal manhattan distance of every origin/destination pair (default 10)
• -a, --origin: origin of a pair to test (format Y:X, has preference over -n and -l)
• -b, --destination: destination of a pair to test (format Y:X, has preference over -n and -l)
• -s, --save: specify this option to save graphical output to ./output folder
• -p, --display: specify this option to display graphical output on a screen
• -h, --help: print out available options

Graph format

Graph is loaded from a text file. Graphs should be specified in a grid format. Format examples are shown in ./test_cases folder.

• . - safe node
• x - fatal node
• (space) - wall

Metrics and strategies

Strategies used to construct the optimal paths for a problem instance are defined in main.py file field tested_strategies. Each OD pair will be evaluated using these strategies and executed against all specified disturbance players (see below). The output is both text and graphical. The text output is stored in raw_results.csv file and its format is defined by the header row. Each consecutive row is then an information about single execution.

Strategies are built using metrics. All metrics for dynamical programming computed strategies are implemented in metrics.py file. To create a new metric just extend Metric interface from this file and insert the resulting class in a DynamicProgramming object. Strategies computed with a Dijkstra algorithm don't require any extra implementation. Just specify your cost function and insert it into new CombinedPathStrategy object.

Disturbance agents

Disturbance agents are used to pick and trigger disturbances during the execution of the strategy. Agents used in the evaluation are stored in file executor.py in a method get_dist_players. Their interface and specific implementations are coded in player.py.