fdotgoodman / genetic-rubbish-collection

This agent-based simulation, implemented through Repast Simphony, looked to incorperate a genetic algorithm to intelligent agent rubbish collection in a spatial coordinate system.

Rubbish Collection Simulation using a genetic algorithm

This project proposes a state-based rubbish collection simulation. In this, the collector agents move around the environment to memorise rubbish locations (MAP STATE). They then use this memory, along with a genetic algorithm to plot the fastest route through the environment to collect all the rubbish (CALCULATION STATE). Finally, the agents action the planned route, collecting rubbish (ACTION STATE) - similar to the Travelling Salesman Problem

The following questions regarding sensitivity analysis are posed:

• How do different heuristics and hyperparameters compare when looking to clean up the most rubbish in a fixed time period? (e.g. hill climbing, crossover, mutation, depth of search, intensity of mutation)
• How does the number of agents affect the speed at which rubbish is picked up, or the number of bits of rubbish that are picked up in total?
• How does the number of bits of rubbish in the environment affect the agent's ability to map out rubbish, or their ability to find an optimal route to collect all the rubbish?

genetic-rubbish-collection's Issues

Do the requisite JavaDocs

Implement method of movement to pick up litter

Fix bug triggering calculation phase twice

When the map phase ends, the finalising of the route occurs twice. Find out why this is and fix it please.

Create Mutation Heuristics

Change all reference of "Genetic algorithm" to "Memetic Algorithm

Incorporate intensity of mutation and depth of search into the heuristics

Add steady-state replacement strategy, and allow the user to choose in the GUI

Create Collector Agent

Implement mechanism to skip over rubbish if it isnt there

This is necessary for multiple agents in case one has already picked up an item of rubbish

Create Phase-Related Parameters

Add a parameter to control the number of ticks before moving to the calculation stage

Write up the report

Create Hill Climbing Heuristics

Create Crossover Heuristics

Create State Transitions

TEST EVERYTHING

A cautionary tale.... Dont just implement everything and hope it all works, only for you to have a bunch of code that you think should work file.

Create Rubbish Agents

GA or MA - Figure it out

Decide on termination criteria for the genetic algorithm

Think about encapsulating all the Collector parameters into a single object

There are a lot of parameters that need to be passed to a collector. Think about encapsulating them all into a single object.

Port the population size of the genetic algorithm to user parameters

Restructure Solution Encapsulation

Ensure that every solution is self-contained in that it can be cloned, has an objective function value, can be mutated etc.

Ensure that SolutionLength is updated correctly at all points, this could cause problems if not...

Figure out what the stopping criteria strategy is, and implement it

Figure out what heuristics will be implemented

Create mechanism of generating an initial population

As part of this, port the initial population parameter to he Parameters User panel

Add memetic algorithm iteration parameter to Repast GUI

This allows the user to specity the number of iterations the Memetic algorithm undertakes as a max length.

Seeing some very weird issues - IndexOutOfBoundsException

Even after debugging, I'm seeing some very strange issues occur. All revolving around IndexOutOfBoundsExceptions being generated in the crossover.

After some preliminary debugging, this might not actually be being caused by this - It is possible, but it looks like at this time, there are duplicate entries of cities to the parent of the crossover. I have my suspicions that the issue could be arising elsewhere.

Create controller

Create GA Structure

Redo Javadocs - Just to make sure they are updated

Think about a smarter strategy for GA termination

As it stands, the termination criteria is simply the number of iterations in the GA. As an extension for this project, it might be a good idea to make it a bit smarter, possibly incorporating population convergeance at a local minima for a certain number of epochs.

Give Collectors a View Distance

Create icons for the collector and rubbish

Add solution representation printing function

Decide on, and implement parent selection process

Redo user parameters - I think the name of the MAP iteration one is not correct, and needs changing

Think about the need for agent-interaction and state based programming

I'm ever so slightly concerned that there is not enough of that DIA aspect here. Have a think about what it needs, and have another read over the brief.

Add mechanism to continue mapping until we find no more rubbish

This effectively adds a stopping point to the simulation, after dormancy, the agents look again and see if they can find more rubbish. If so, do the MA again, if not, go to sleep

Figure out and Implement MAP_STATE Movement

Move randomly? Or make get random position and move to it?

Carry out literature review on the solution

Decide on, and implement generational replacement strategy