Hello ,
now I really have to thank the OP for opening this thread! It is really great of you to contribute here! Needless to say that we knew your name already!
Can we interview you a bit?
In your Jacobian Transpose example there are, depending on the complexity of the upper body, a minimum of 4 joints from hand to pelvis (wrist, elbow, shoulder, neck). So the Jacobian would contain 4 equations that would have to be iteratively solved. (hope that is right). How many iterations are typically required for a result like in your 2016 video (Guided Learning of Control Graphs for Physics-Based Characters).
The locomotion āfeelsā very natural in this video. It is one of the most overlooked field in games, as it seems, because not long ago CPUs were just to weak. Now that i-5 and i-7 generations populate the gaming world, the field is open for physics based games. Doing things physics based adds just so much immersion to a game because you can instantly connect to the character based on your physical experiences from real life. It is really strange that only the GTA series uses it extensively with Natural Motion integration.
Wee love that topic.
I tried the countermassstuff too.
Normally i would add a āKreiselkompassā/gyro, for stabi, but in realworld i would build it with my hands.
Inside ue4 i need my brain.
I am lost. ^^
This topic is awesome. Iāve long had an interest in trying to use a physical character controller for a game, but itās always seemed just out of reach.
Itās an especially good time for this given that we have access to the full PhysX core source code in UE4.
Sir jumpbot needs rollerblades and some thrusters now.
Funstuff ^^
I did the same with the āheadā of jumpbot.
āHeadā is hollow object. I used a ball/sphere as ābrainā (negative Z)to check stuff, but velocity has to be smoothed out to much. No āquickburstā possible, without glitching.
Edit: Now i try to replicate those legs with physics, because i think those are one of most interesting legs, nature ever built.
Not much time next days, but i want to figure it out. ^^
www.scienceinschool.org/sites/default/files/articleContentImages/21/ostrich/issue21ostrich5_xl.jpg
Hey , itās great to see you here and thanks for releasing the cartwheel source code. I love your site, itās definitely the best place to find information about articulated characters. The 2016 papers are amazing, particularly the āGuided Learning of Control Graphs for Physics-Based Charactersā paper. It would be incredible if Unreal had similar plugins. I look forward to seeing your future projects.
I was actually thinking about trying the stable PD controllers back when I was testing Bullets Featherstone implementation in Unreal. I didnāt know that Unreal included the PhysX source code, Iād prefer to use PhysX if possible as itās already integrated. I might add it to my TODO list. The Jacobian Transposes seems really useful, I should probably study it a bit more.
The counter weight concepts that BlueBudgie and Luftbauch are trying seems like an interesting problem. It reminds me of a high-wire robot I once saw.
I will definitely try next week to understand the one leg hopper. Also exciting to see the leg structure of the bird in the other post.
In case you have access to some Matlab (with Simulink+SimScapeMultiBody) license, here is a good project. We used Matlab quite a bit to prototype things for R.C. Bot Incā¦ It is quicker than C++.
www.mathworks.com/matlabcentral/fileexchange/21439-simulation-of-a-one-legged-hopping-robot
Raibert also seems to have published something on the MIT servers, but it did not open.
The hopper is interesting, because it would make a fun game that is in reach when you are only a handful or a dozen of people. Like āhopper battlesā in arenas. They could even have little physics cannons or things. The pure slapstick alone would keep people entertained.
It feels generally much more realistic to make a fun physics based game when you are that size instead of dreaming of creating huge open worlds in UE4 that typically requires hundreds of professionals and years to make. Thatās just unrealistic.
So we looked into the one leg hopper and it is a bit disappointing. It turns out there is no analytical solution to it. Best summarized here
www.ritsumei.ac.jp/se/~gen/Passive/passive_en.htm#one-legged
With such a physics hopper in multiplayer you would either have serious load if you calculated all hoppers on one server and distribute the simulation results to the clients, or you would get different results over time if you did the calculations on each client.
You also could not interpolate frames or do other optimizations because accumulated deviations from the exact solution could build up and spiral out making the hopper get messy. The balancing is just too sensitive.
It is something where you know it can be done, but you also know in real life it will be unusable. It seems to be something where you also better do not ask for performance figures and is best suited for academia to get students practice time, but we have grown out of that pair of shoes. But for academia it might be useful when you are on your knees begging for funding. It is interesting after all.
