That's a really cool project! And it made me want to play Quake again.
Show HN: Quake 1 movement physics reinforcement learning project
11–20 of 21 posts
Re: Show HN: Quake 1 movement physics reinforcement learning project
#12Prepare for something awesome
Human run [1] vs scripted bot run [2].
This is from DFWC, a race series that takes place every second year, and using modified Q3 engine.
There are 5-7 maps per season, and people race from start to finish, and best time wins.
Also, each round there is scripted run, to show how far you could push the engine physics, if you had inhuman reflexes. Highly recommend browsing entire YT channel, it's really awesome, if you ever played Q3!
Re: Show HN: Quake 1 movement physics reinforcement learning project
#13Given the fully trained RL model, would it be possible to infer and explain the optimal movement technique in terms of a simple rule set?
Re: Show HN: Quake 1 movement physics reinforcement learning project
#14Now do CS:GO!
Re: Show HN: Quake 1 movement physics reinforcement learning project
#15Prepare for something awesome Human run [1] vs scripted bot run [2]. This is from DFWC, a race series that takes place every second year, and using modified Q3 engine. There are 5-7 maps per season, and people race from start to finish, and best time wins. Also, each round there is scripted run, to show how far you could push the engine physics, if you had inhuman reflexes. Highly recommend browsing entire YT channel…
I don't know what I just watched, but these movements seemed not very physically accurate to me.
Re: Show HN: Quake 1 movement physics reinforcement learning project
#16Prepare for something awesome Human run [1] vs scripted bot run [2]. This is from DFWC, a race series that takes place every second year, and using modified Q3 engine. There are 5-7 maps per season, and people race from start to finish, and best time wins. Also, each round there is scripted run, to show how far you could push the engine physics, if you had inhuman reflexes. Highly recommend browsing entire YT channel…
I don't know what I just watched, but these movements seemed not very physically accurate to me.
The goal is to push the limits of the game's physics not real life. Quake based games allow you to retain speed when jumping at the same 'frame' as landing (bunny hopping) and slide up slopes with enough speed.
These physics create a high skill ceiling for speed runs and in normal play because of how powerful movement becomes during play.
Re: Show HN: Quake 1 movement physics reinforcement learning project
#17Karl Jobst has a Youtube video covering the speedrunnning record history of Quake's first level, which serves as a good overview of Quake movement technique:
https://www.youtube.com/watch?v=43d8fICz6gM
It starts with an explanation of the significance of Quake to speedrunning as a whole, so skip to 5:46 if you only want to see the movement tech.
Re: Show HN: Quake 1 movement physics reinforcement learning project
#18Karl Jobst has a Youtube video covering the speedrunnning record history of Quake's first level, which serves as a good overview of Quake movement technique: https://www.youtube.com/watch?v=43d8fICz6gM It starts with an explanation of the significance of Quake to speedrunning as a whole, so skip to 5:46 if you only want to see the movement tech.
This is an incredible video - thanks for sharing. Couple notes for those like me who are new to the speed running world:
* The goal here is to get through a particular level as fast as possible. Not to kill enemies or get points; it's just speed.
* The physics engine computes a speed for your player. There are some movements that increase it and some that decrease it. What these speedrunners are doing are (1) finding techniques in movement (2) exploiting features of the map that result in increasing their speed and maintaining it.
* These techniques include running along a wall (unclear to me why this increases or maintains speed), bunny hopping (jumping and turning at the same time in such a way where you don't pick up friction), leveraging features in a map to give you added speed (blowing up an explosive barrel and using its shock to speed you up).
* Over the course of 20 years, players have managed to get the record from 30 seconds down to its current 20 seconds for the quake map in question.
As someone who spent many hours playing Quake 2 on multiplayer, the best players moved in a rhythm similar to how the speed runner players do here.
Re: Show HN: Quake 1 movement physics reinforcement learning project
#19Karl Jobst has a Youtube video covering the speedrunnning record history of Quake's first level, which serves as a good overview of Quake movement technique: https://www.youtube.com/watch?v=43d8fICz6gM It starts with an explanation of the significance of Quake to speedrunning as a whole, so skip to 5:46 if you only want to see the movement tech.
This is an incredible video - thanks for sharing. Couple notes for those like me who are new to the speed running world: * The goal here is to get through a particular level as fast as possible. Not to kill enemies or get points; it's just speed. * The physics engine computes a speed for your player. There are some movements that increase it and some that decrease it. What these speedrunners are doing are (1) finding…
Speculating here based on my experiences in the past building my own physics engines: running along a wall could speed you up a little bit depending on how they’re handling “keeping you outside the wall”. One common/simple way is to apply x1 = x0 + v*dt, calculate collisions after the moment calculation, and then fudge your position back to being inbounds. If I recall, Q1 used simple bounding boxes on all of the entities as well and there’d probably be some (pardon the pun) corner cases in the “keep out of the wall” algorithm.
Re: Show HN: Quake 1 movement physics reinforcement learning project
#20That's a really cool project! And it made me want to play Quake again.
http://fte.triptohell.info/demo
There ya go!