Steering evaluate fitness

Question

I've made a simple game with a steering model that manage a crowd of agents. I use an genetic algorithm to find the best parameters to use in my system but I need to determine a fitness for each simulation. I know it's something like that:

number of collisions * time to reach goal * effort

But I don't know how to calculate the effort, is there a special way to do that ?

Here is what I've done so far:

// Evaluate the distance from agents to goal
Real totalDistance(0.0);
for (unsigned i=0; i<_agents.size(); i++)
{
    totalDistance += _agents[i]->position().distance(_agents[i]->_goal->position());
}

Real totalWallsCollision(0.0);
for (unsigned i=0; i<_agents.size(); i++)
{
    for (unsigned j=0; j<walls.size(); j++)
    {
        if ( walls[j]->inside(_agents[i]->position()) )
        {
            totalCollision += 1.0;
        }
    }
}

return totalDistance + totalWallsCollision;

Thanks for your help.

Answer 1

Effort will be specific to your simulation. You could do something like the sum of the distance each agent traveled. Or the sum of all the direction changes each agent had to make. It depends on what your definition of fitness is.

For either of the two above, you can just keep a running total for each agent, then add them all up, just as you do for distance:

Real totalEffort(0.0)
for (unsigned i=0; i<_agents.size(); i++)
{
    totalEffort+= _agents[i]->totalEffort();
}

return totalDistance + totalWallsCollision + totalEffort;

You could also try keeping a running total of work, so each step of your simulation, calculate aceleration * mass * distanceTraveled. Where distanceTraveled is the distance traveled since the last update. This will give you the total work (effort) that the agent exerted.