AI Is Forced To Play Table Tennis (Part 2)

if you watch part one then you know what's about to happen but if not let me fill you in so far we've created two systems a walking AI with hand control and an a-bot for table tennis and now it's time to combine them together this is no easy task you can think of this as an Unholy marriage between two people who have nothing in common there will be a lot of issues but maybe just maybe if they work together then they might create something beautiful let's see if we can solve these issues the paddle will be held in the ragdoll's right hand and while the AI has some pressure control over it there is a limit to how far it can reach so our aimbot's output must stay within Arm's Reach of the shoulder the a bolt also requires a set location to plan a strike again the ragdoll's movement's kinda limited so the AI may not be able to reach it instead we will constantly update the interception point to be half a meter in front of the Ragdoll this does restrict the type of shots it can hit but it will also help the Aimbot suggest shots to the AI that it can actually form finally we will allow the Aimbot to control the angle at which the paddle points normally I would give the AI Control over this but I think it's a bit redundant there's no dynamic system in the wrist it simply snaps to whatever angle it's given so there's actually nothing for the AI to learn the things we've done to our Aimbot have now made it Ragdoll friendly but we still need to get our Ragdoll to be Aimbot friendly to get our Ragdoll to cooperate with the Aimbot we will have to do two things first we will set its walking Target to be just behind the table this will keep the Ragdoll close to the action and allow it to be flexible in dealing with shots second we will set its hand Target to the aimbot's output we already know that the Aimbot suggestions are safe for the Ragdoll to use so he can be fairly confident that the AI will be able to carry out the shots at once and surprisingly that was the last thing we needed to do our marriage was successful and our table tennis player has come together so all we need to do now is clone it apply a new paint job and we can start watching a virtual table wait what fun fact about my AI it can walk perfectly when it's facing forward but when it's turned around it collapses in on itself like a dying spider now better programmers than me would probably cut their losses and just retrain the AI but honestly I can't be asked what I'm going to do instead is Trick the AI into thinking it wasn't rotated by flipping its inputs back there are two types of inputs we need to flip vectors and quaternions we can flip vectors 180 degrees by negating the x and z axes which is fairly straightforward quaternions are a little trickier the easiest way to flip one is to multiply it with another quaternion which already has that 180 degree rotation so I did that and lo and behold it didn't work it was at this point that chat gbt informed me that there are eight different ways a quaternion can represent the rotation I had my dumbass AI was only aware of the first way so when it encountered any of the other forms it crumbled I soon discovered that the quaternions I had were in the seventh form so all all I had to do was negate the Y and W axes to revert it back to the original this simple tweak solved the problem and the blue AI started working perfectly we've now run out of things to talk about and there's two highly skilled AI in front of us itching for a challenge so let's set them against each other in a best of 21 match [Music] laughs foreign foreign foreign foreign foreign [Laughter] foreign foreign foreign thanks foreign thank you foreign foreign foreign

Channel: b2studios

Views: 82,171

Keywords: b2studios, ai, b2studios ai, tabletennis, tabletenis, ping pong, b2studios ping pong, ping ping, b2studios tabletenis

Id: WS9ST9A6XEM

Length: 12min 23sec (743 seconds)

Published: Fri Aug 11 2023