Simplified Rover Simulation using JAM Intelligent Agent Framework and Python for Visualizations (matplotlib) 🤖
This repo presents a simulation of a simplified rover whose primary goal is to explore the rocks placed in a an two dimensional grid. The rover has a number of capabilities i.e. moving around the area, analyzing a rock to check if it has traces of water and taking it back to the base if it does. The problem can be expressed as follows; given a m×n grid with total area α and r rocks placed in certain coordinates, we need to ensure the rover travels to the locations where the rocks are, analyzes them and brings them back to base (x0,y0) if they have traces of water w. The rover is controlled by an Intelligent Agent to direct the movements and actions. This agent was developed using JAM Intelligent Agent Framework and several tests have been conducted to evaluate it’s performance in different conditions.
The overall simulation system is composed of three components; Dynamic World Model Generation (Facts), JAM Agent and Simulation Visualization. The entire system can be run solely using the run.py python script; main() method handles the execution of all components.
This component allows the user to dynamically generate different world models file that the JAM agent can later be run with. Both user-entered input and random generated samples options are supported. When entering the data the user can specify state variables such as base position, starting position and rock locations. The generated model data file is stored in the experiments folder with the filename of format; facts_n_m_n.txt. This component is used to setup different conditions for running experiments and evaluating the agent behaviour in those conditions. The following methods in the run.py file make up this component; getInput(), genRandom(), CreateJAMFactsFile().
This component allows the user to run the jam agent with the specified world model. The runJAMAgent(facts_filename) method in run.py file takes in the the dynamically generated world model filename and invokes mars_rover_agent.jam with correct jam.jar java package distribution in the command prompt. The jam agent logs each action to the command prompt stdout. Once agent finishes, this stdout stream is then returned by the method for further use in the overall system and a file with same world model filename + '_output' is created in the same directory where the world model text files exists. The agent has been designed in a way that it works for different rock locations and different grid sizes (including negative axis) i.e entire 2d space.
The core JAM agent design includes the following facts, goals and plans:
These are passed in from the facts file and not directly included in the agent code, this is so that we can run same agent code for multiple world models. Currently the agent interacts with the following facts during execution:
- Base Location - this store the coordinates of the rover's base location in the grid.
- Number of Rocks - this tells the agent how many rocks there are in the world model so that it can go to analyze them.
- Rocks - this stores information about the coordinates of the the rocks and it's features.
- Rover Position - this stores information about the rover's current position coordinates as the rover is moved around the area α.
There are two types of goals set; top-level goals and sub-goals. Each goal also has a plan included i.e the steps the agents needs to take in order to successfully achieve the goal. The following are the details:
- Explore Rocks - this is our top-level goal telling the agent to explore all the rocks from the world model.
- Go to Position - this a sub-goal in top-level goal and is achieved when the agent successfully goes from position a to b. It has further subgoals such as moving forward, backward, left and right.
- Analyze Rock - this goal is used when it finds a rock and then it looks to for traces of water.
- Pick Up Rock - this goal is used when the agent finds a rock with water, it removes the rock from that location and marks the rock as found.
In order to be able to monitor the agent better, a visualization component has been added. This component is written in python and uses matplotlib package for plotting the grid, rock locations and rover movements. The idea is that, once the agent finishes running the outputs from the stdout is parsed into appropriate data structures that we can further use for plotting. The methods parseJAMOutput and vizualize in run.py are responsible for handling the parsing and visualizing of the agent output. Moreover, there is also a replay method, which can take in the agent output file and run the visualization of that particular simulation. It is helpful when certain simulations take too long to run from scratch, with this in place we can just run it once and visualize it multiple time to evaluate the agent behaviour.