2 hours ago, JoeJ said:

You can fake graphics, but not physics.

Of cource. Becouse gaming collision response involve to mach black magic.

16 hours ago, JoeJ said:

n that case feet contacts form a cycle and calculating a solution becomes hard

I know result exactly. It  will complitely outclass  human on task known as "make a higher posible volume of BIG-BADA-BOOOM sound" becouse it human-like. 
 

1 hour ago, JoeJ said:

But what if it is a walking robot standing with two feet on the ground

Human and human like-machine is staticaly unstable becouse it have only 2 support points and upper centre of mass. It have to jit anytime to keep balance dynamicaly. It is why imposible to sleep standing and why drunk guy's jittings have a very high magnitude - his response is to slow to keep balance accurate. Just get one you leg up  and close your eyes. You will feel jitting by foot of other leg.Closing of eyes just turn off narrow position feedback sensing so you can feel jitting amplitude increased. So real robot will anytime seek stable vertical position but can not reach it, so jitting around it like human. Upper mass balance is unstable. Also it just unable to measure boxes properties that required to simulation.

Just imagine that you have a compass and goes directly to the North Pole. Each step you go into direction that compass needle shows. It how any gradient climb solvers works (both digital and analogue, including many human brain functions) - gradient show direction of shortest way to function extremum .  But when you approached close to North Pole you step length exids exactly length that needed to make step directly to Pole. So you step over a Pole and now compass needle show opposite direction. You step again and it again exids needed  length. It is jitter effect come from. It why unregulated actuator jitting around target point. To fight jitting effect you have to decrease step proportionally to your distance from Pole. It how any actuators works. But any mechanics (including human muscles) have minimal step length limit. To fight it limit you have to reach target point with given tolerance, that exids minimal step limit, and stop actuator. But you can not stop balancing a torso, becouse any error will be increased due to gravity.  Balancing of torso causes jitting of actuator joined  to it around target point. To fight it you have to jit actuator respectively to torso. It is why any precious work human have to made sitting on chair and have his elbows resting on supports - torso and legs just is to bad support for precious actuators movements.

Also it is why human-like robots is antiscience fiction that just become to fashion toy now. To extend human's abilities robot must have non-human like construction and brains, or better say autopilot(  Intelligence is really bad thing for driving screws on convwyor and other similar works). So any nonstationar robot must have at least 3 support points and low profile to be staticaly stable and have ability to spend accumulator energy to solve tasks that it intended for,  instead for  play to balancing games. Just look at this video https://www.youtube.com/watch?v=z-mub6RXzAU.  Actuator have same kinematics as human's hand. It clearly shows what support area and mass needed to get only actuator stable and what happends in case stand's support  area is  not enought. Couple days ago i has seen similar machine that has pulled out from ground 2-meters  height concrete pillars together with meter of underound concrete base like mushrums. To done something like this human-like robot require size at least as Lord Megathron. Only task of real robotics is to make industrial machines and industrial processes unmaned, to liberate human from slavery of machines.

On 6/19/2018 at 3:06 PM, Fulcrum.013 said:

Clusters of  particles is not a neural networks at all. It is tool that allow to mix efficiency of Euler grids approach with flexibility of Lagrange volumes approach for integration of finite difference schemes. Also it makes collision detections friendly to GPU by replacing complex curved  surfaces, wich collision detections about imposible to vectorize, by sets of multiple spheres, wich collision detection is perfectly vectorizable on GPU cores.   

I do not understand what you said. Your response is incoherent. Grammatically incorrect and logically inconsistent with the conversation. I don't think English is your first language. 

From what I understand about neural networks, they're good at "guesstimating" functions for things.
I wonder if it could guesstimate something for a cluster of particles.

2 hours ago, Fulcrum.013 said:

Human and human like-machine is staticaly unstable becouse it have only 2 support points and upper centre of mass.

Ha, my 'spare time expertice' ... can't resist to show my walking ragdoll again: https://www.youtube.com/watch?v=ULRnlAbtL3s

The video is old and looks robotic (upper body does nothing, dump state machine to drive walking), but i hope i have time to continue on this in the future. Luckily Newton engine cares for all the constraint solving stuff we've discussed, so i can focus only on the control problem.

Balancing is really a critical action. Humans do it always being at the edge of what is physically possible. Took me years to figure out how it works. For the simulation i do it analytically on an inverted pendulum model, which i have to map to and from the ragdoll bodies. Finally i set target angular acceleration to the joint motors, and Newton cares for the rest. No cheating here and performance is game ready.

If' i'll ever make a game on my own again, then only with this kind of characters - otherwise i won't! 

3 hours ago, JoeJ said:

The video is old and looks robotic (

System can have a 2 points of balance - upper and lower, like a flip-flap. It can be found upper  point where it balanced, but physic body anytime seeking a lower balance point. So to keep upper balance bipedal walker have to make a work against gravity force anytime, while on low-balanced systems grawity works to keep it balanced. It why homan-like walkers is nonsence into industrial robotics field. But it can be a very intrasting toy.

Also human walks by other scheme. Human  moves a torso litlle bit forward to lose bnalance , and allow potential energy accumulated into upper balance to move mass center forward, than catch balance again by steping one leg forward, and restoring energy moving centre of mass up by forward leg while moving back leg forward.  Really until 20-th century russian army has a command "Move right shoulder forward" just prior "Step march" command.

45 minutes ago, Fulcrum.013 said:

System can have a 2 points of balance - upper and lower, like a flip-flap. It can be found upper  point where it balanced, but physic body anytime seeking a lower balance point. 

Can you give real world example? Googling flip-flap only gives me children books.

Early on i believed that hip movement has some role in balancing, and it might be the reason why i was able to swing left and right while keeping stable balance much faster than my ragdoll was able to do. But after understanding the inverted pendulum i solved this and ragdoll can swing as fast as i now. I concluded that only the center of mass and support polygon relates to balancing, nothing else (assuming static friction is large enough to prevent any sliding, and you do not make a flywheel from swinging arms while tipping back over an edge etc.).

This implies that all the balancing we do comes from the ankles, and other internal motion is basically a free variable. In fact i can play back some animation on the upper body and the controller adjusts the ankle torque automatically. So yes, eventually those 2 points balance you talk about could help me to make this internal motion look natural...

5 hours ago, NoAbsoloops said:

From what I understand about neural networks, they're good at "guesstimating" functions for things.
I wonder if it could guesstimate something for a cluster of particles.

(First, sorry for the off topic stuff.)

One of its pioneers said that actual machine learning is just very advanced curve fitting. There is not so much hope that this is faster than just calculating those curves from physics equations directly, i guess.

But i don't know. What you ask for is no longer physics. It might be the wrong forum, or even the wrong homepage, or just too early to ask for this. I wonder Nvidia has not already tried something in this direction. We'll see probably...

Anyways, what you want has not yet been done in games i guess? So it is probably very hard to achieve. Not knowing what you aim for in more detail makes it hard to even make an estimation if it might be possible or impossible.

5 hours ago, JoeJ said:

I wonder Nvidia has not already tried something in this direction. We'll see probably...

Just neural networks and "deep learning" is other antiscientic fiction fashion toy. Really neural networks intended to determine category of forms for wich exactly determination rules can not be found. It is what neural networks making really good. But neural networks can not make a exactly desigin by nature of task for wich it intended, its can give estimations only. Of course it is not apropriate way for realistic phisic simulation.

5 hours ago, JoeJ said:

Can you give real world example? Googling flip-flap only gives me children books.

It is a inverted pendulum bench used to train paratroopers etc, also known as looping.

5 hours ago, JoeJ said:

Early on i believed that hip movement has some role in balancing,

Hips used to lift torso up and control descending of torso when human climbing up or goes down ladder/hill.

5 hours ago, JoeJ said:

This implies that all the balancing we do comes from the ankles,

Exactly. We exactly know that human able to stay onto one leg. To transfer its stable system to other location we have to make longer other leg.  It can be done by rotaing ankle joint only. By other worlds we can assume that working leg is crane's jib and rest of body is a load. By the way human kinematics allow significant simplification. Hands and legs is a icosceles triangles. So elbow anytime can be found at  middle normal to  line connecting shoulder and center of palm, same with knees. So we can consume that arms and legs is a linear joint  connected to angular joint of torso on general pose computations, and then recalulate positions of angular joints of each hand and leg separately.