We need to map out an unknown ship and find the shortest path the oxygen generator. However, we're only able to explore the ship by issuing commands to a robot controlled by intcode. The robot has no distance vision; to know where a wall exists, you have to bump into it.
My hasty approach when solving the problem was to issue random walk commands until the entire maze was mapped out. This worked, but it was unneccessarily slow: it took about a minute or two to finish. After the entire grid was mapped out, I used BFS to find the shortest path to the generator. I didn't use BFS for the initial exploration because it seemed difficult, since all exploration was done with a stateful robot that I had to send individual movement commands to. I couldn't just move to arbitrary squares as the algorithm would need me to do.
I tried tweaking the random walk by biasing it in certain directions, which rotated, or having the robot bias towards unknown squares.. but these small improvements didn't help that much. The algorithm was fundamentally random and thus, slow, since it would often do repeated work. This is a deeper lesson I've learned this AOC. Making small tweaks upon a fundamentally incorrect algorithm doesn't help much, and it's easy to lose a lot of time here. It's often better to throw away the algorithm and use a different one. That's where huge gains are.
A few days later, I went back and reworked the exploration algorithm to use DFS. For each location, I stored the state of the Intcode VM's memory, then before issuing a movement command I would restore the state. This meant that even though the recursion would eventually lead to big gaps in exploration (issue the last command from square 20,20, find nothing, then pop back to 15,15 in the tree and search there), it was no problem. Instead of taking a minute or two to explore, it was done in a second or so. Much nicer!
I was also able to remove the separate search for the oxygen sensor after the maze is mapped out. The DFS records path information, so I just use that to find the path needed in Part 1.
Part2 asks how long it takes for oxygen to fill all of the maze. I used a simple flood fill, scanning over the maze, marking the squares that would be filled with oxygen next, adding it to the marked squares and repeating. The key here seems to be not mixing your scanning and adding phases. Another approach would be to find the longest depth from the oxygen generator using BFS/DFS.
2019 Day 15 on AdventOfCode.com
Out here in deep space, many things can go wrong. Fortunately, many of those things have indicator lights. Unfortunately, one of those lights is lit: the oxygen system for part of the ship has failed!
According to the readouts, the oxygen system must have failed days ago after a rupture in oxygen tank two; that section of the ship was automatically sealed once oxygen levels went dangerously low. A single remotely-operated repair droid is your only option for fixing the oxygen system.
The Elves' care package included an Intcode program (your puzzle input) that you can use to remotely control the repair droid. By running that program, you can direct the repair droid to the oxygen system and fix the problem.
The remote control program executes the following steps in a loop forever:
- Accept a movement command via an input instruction.
- Send the movement command to the repair droid.
- Wait for the repair droid to finish the movement operation.
- Report on the status of the repair droid via an output instruction.
Only four movement commands are understood: north (1), south (2), west (3), and east (4). Any other command is invalid. The movements differ in direction, but not in distance: in a long enough east-west hallway, a series of commands like 4,4,4,4,3,3,3,3 would leave the repair droid back where it started.
The repair droid can reply with any of the following status codes:
- 0: The repair droid hit a wall. Its position has not changed.
- 1: The repair droid has moved one step in the requested direction.
- 2: The repair droid has moved one step in the requested direction; its new position is the location of the oxygen system.
You don't know anything about the area around the repair droid, but you can figure it out by watching the status codes.
For example, we can draw the area using D for the droid, # for walls, . for locations the droid can traverse, and empty space for unexplored locations. Then, the initial state looks like this:
D
To make the droid go north, send it 1. If it replies with 0, you know that location is a wall and that the droid didn't move:
#
D
To move east, send 4; a reply of 1 means the movement was successful:
#
.D
Then, perhaps attempts to move north (1), south (2), and east (4) are all met with replies of 0:
##
.D#
#
Now, you know the repair droid is in a dead end. Backtrack with 3 (which you already know will get a reply of 1 because you already know that location is open):
##
D.#
#
Then, perhaps west (3) gets a reply of 0, south (2) gets a reply of 1, south again (2) gets a reply of 0, and then west (3) gets a reply of 2:
##
#..#
D.#
#
Now, because of the reply of 2, you know you've found the oxygen system! In this example, it was only 2 moves away from the repair droid's starting position.
What is the fewest number of movement commands required to move the repair droid from its starting position to the location of the oxygen system?
You quickly repair the oxygen system; oxygen gradually fills the area.
Oxygen starts in the location containing the repaired oxygen system. It takes one minute for oxygen to spread to all open locations that are adjacent to a location that already contains oxygen. Diagonal locations are not adjacent.
In the example above, suppose you've used the droid to explore the area fully and have the following map (where locations that currently contain oxygen are marked O):
##
#..##
#.#..#
#.O.#
###
Initially, the only location which contains oxygen is the location of the repaired oxygen system. However, after one minute, the oxygen spreads to all open (.) locations that are adjacent to a location containing oxygen:
##
#..##
#.#..#
#OOO#
###
After a total of two minutes, the map looks like this:
##
#..##
#O#O.#
#OOO#
###
After a total of three minutes:
##
#O.##
#O#OO#
#OOO#
###
And finally, the whole region is full of oxygen after a total of four minutes:
##
#OO##
#O#OO#
#OOO#
###
So, in this example, all locations contain oxygen after 4 minutes.
Use the repair droid to get a complete map of the area. How many minutes will it take to fill with oxygen?