Machine Learning Dev --> Waypoints for Rl-Agent

[Gizmotronn]

This milestone is for Machine Learning Development for <AC0/RD> applets

#88 #89 #90 #91

@Exynos-999/AWS_JPL_OSR_DRL --> slack/11.3.20, 12.3.20

[Gizmotronn]

Were we able to get anywhere with the new waypoints scripts?
not yet . i was finishing a few tests from yesterday, it makes through to waypoint 5 ,a few times during the sim. i going to try to space out the waypoints a little further . waypoint 5 is still about 8 meters from the checkpoint through a lot of rocks.

Liam ARBUCKLE:slack_call: 15 minutes ago
Would you like me to do that? Can you send through the coords for the initial position of the rover and the checkpoint? Are we spacing them out in 5ths, or 3rds - I think 3rds would be better, as it means that it doesn't direct the rover in a straight line too much

Liam ARBUCKLE:slack_call: 11 minutes ago
How much does the rover attempt to follow the waypoints? In the actual test by the judges would the waypoints be kind of close to potential obstacles? Big problem is that waypoints may direct the agent too much, and we place too much reliance on them. Since we can't control where the WPs will be in the actual sim, we could try to connect the pathfinding models to the waypoints, and the rl-agent sets the waypoints dynamically.

Edwin M 11 minutes ago
currently they are not quite evenly spaces. i started with even spacing, and the rover didn’t make it to waypoint 1. so i moved them a little closer. basically 44 /1.3, 44 / 1,8, 44/2 , 44/3, 44/4. but this was before i added a smoother reward function for distance. Perhaps I’ll try thirds again.

Liam ARBUCKLE:slack_call: 9 minutes ago
Coords for the init position of agent? Final checkpoint? I could quickly make a commit in my branch w/out having to search through the scripts to find these. I have about 5 minutes before I have to go, so I'd need those now.