@Gargomon251 Yeah, I can't remember if this was a limitation of the API at that time or if I was just too busy implementing other parts of the AI. For example I didn't even tried to extract data about the coins from the API because it wasn't important for the competition. It was impossible to retrieve Marios state (small, big, white..) so I didn't even tried to make it handle fireballs. I made the video to visualize the restricted reconstruction of its world my controller used for its decisions.
@amoxys That is incorrect. There were/are methods to retrieve mario's state AND differentiate between different elements. See my Mario controller for an example of one that does that. -Erik
@WolfCoder you're right. but on the right side of the video (my agent's perception) there is no gate. as i can remember there was no way to retrieve the position of the gate.
Thank you! But it isn't even aware of holding the shells. It just keeps pressing the run/hold button all the time. Even worse: It used to run away from shells it was holding. :D I still haven't implemented a routine to detect if it is holding something. But it ignores unavoidable collisions now.
Yes, the agent estimates the trajectory to avoid falling into holes and bumping into enemies. You can see how he refuses to finish the level because all he sees is a hole.
I can't help but notice it doesn't detect mario's fireballs or coins, nor can it tell mushrooms from enemies.
Gargomon251 1 year ago
@Gargomon251 Yeah, I can't remember if this was a limitation of the API at that time or if I was just too busy implementing other parts of the AI. For example I didn't even tried to extract data about the coins from the API because it wasn't important for the competition. It was impossible to retrieve Marios state (small, big, white..) so I didn't even tried to make it handle fireballs. I made the video to visualize the restricted reconstruction of its world my controller used for its decisions.
amoxys 10 months ago
@amoxys That is incorrect. There were/are methods to retrieve mario's state AND differentiate between different elements. See my Mario controller for an example of one that does that. -Erik
eerikk 3 months ago
Wow I can't even begin to understand how you did this, I like intelligent people.
Krondelo 1 year ago
This has been flagged as spam show
I completely agree. I just downloaded this mp3 at mp3thatvid..net
joe230smith 1 year ago
it doesn't need to handle the abyss if jumping into the "abyss" makes mario hit the gate
WolfCoder 1 year ago
@WolfCoder you're right. but on the right side of the video (my agent's perception) there is no gate. as i can remember there was no way to retrieve the position of the gate.
amoxys 1 year ago
You must be a very clever man
ert135 2 years ago
@ert135 yep :)
amoxys 1 year ago
That's really cool! I like the way it deals with shells.
DirtbagD 2 years ago
Thank you! But it isn't even aware of holding the shells. It just keeps pressing the run/hold button all the time. Even worse: It used to run away from shells it was holding. :D I still haven't implemented a routine to detect if it is holding something. But it ignores unavoidable collisions now.
amoxys 2 years ago
@amoxys XD It ran away from the shells it was holding?
leafbarrett 1 year ago
Very cool; I see you're projecting forward to where Mario will land. Is this your agent playing or you playing by hand? It's kinda hard to tell.
0xec1ec71c 2 years ago
Yes, the agent estimates the trajectory to avoid falling into holes and bumping into enemies. You can see how he refuses to finish the level because all he sees is a hole.
amoxys 2 years ago