Quantify Results

[JohnHaTrick]

We need a way to quantify the progress of our methods.

Maybe compare it to the score of a human player? Compare Forward Search and another method?

[huevosabio]

I think comparing against another method would be nice. We can likely implement two more if we choose them carefully

[JohnHaTrick]

The scoreboard gives us an easy way to quantify the performance.

[JohnHaTrick]

Expected time to accident?

[JohnHaTrick]

Plot rewards and actions in nice formatting.

[huevosabio]

I would like to include a way to spit out reward, and action history into csv

[JohnHaTrick]

MATLAB struct

AA228MDP_uniqueIdentifier.mat

struct with the following fields:

absoluteStates

agent's and obstacles' absolute (not relative) states in the following matrix:
dimensions: (numObs + 1)*simPeriods x 3
each row: [ x, l, v ], starting with the agent

actions

rewards

? simPeriods ( always 1200 for now), others?

.csv files and descriptions

absolutesStates.csv

description: agent's and obstacles' absolute (not relative) states in the following matrix:
dimensions: (numObs + 1)*iterations x 3
each row: [ x, l, v ], starting the agent

actions