Assets APE - DeepMotion Avatar Physics Engine for Robust Joints and Powerful Motors

APE Poser - Inverse Dynamics based Physics Animation Editor

Creating a physics based animation controller for a spider that walks and interacts with the environment in a physically realistic way is not an easy task. Traditional workflow for achieving a physics based animation controller involves the following steps:

1. Rig and animate the spider in an external tool such as Maya or Max
2. Import the rig to Unity and convert it to physics based joints
3. Import the animation to Unity and convert it to physics poses in the physics based joint space
4. Run the physics simulation and check if the kinematic poses react to environment correctly
5. Due to the disconnection between a kinematically animated rig and its dynamic simulation it's likely that the poses need heavy tweaks to produce a plausible physics based animation controller. Which means we need to loop back to step 1 for many times

As we see this is a painful workflow and it is hard to iterate with. Alternatively programmers can "guess" the physical poses and hardcode the joint parameters into the physics animation controller and perform a lot of try and errors without direct visual feedback. Neither approaches is appealing to physics based game developer.

APE supports a very simple but powerful Inverse Dynamics based Physics Animation poser. In the following I will run a quick tutorial on posing the front legs of the the spider's walking controller to illustrate how Inverse Dynamics based posing with APE works.

Let's say we're designing a 4-phase walk cycle for the spider. Due to the symmetry we only need to create two unique poses and let the animation controller to create the mirrors automatically. Here are the steps to do it:

Step-1: Create the first pose that is consisted of a raised left leg and a right leg sweeping back on the ground. As shown in the following video we just start APE simulation in the posing mode and use a physics based dragger to position the legs the way we want. Effectively we are doing ID (inverse dynamics) based posing. ID posing has all the benefits of IK posing plus its unique advantage that ID will generate physically plausible poses obeying the Newtonian laws while IK ignores the mass of rigid bodies and often yields physically impossible poses.

Step-2: Copy the pose out of the position sensors on each DOF of the joint motors in the upper legs and lower legs. Since the upper leg has a ball joint of 3-DOF we need copy 3 Euler angles to the spider controller script. The lower leg is a hinge joint of 1 DOF, thus we only need copy one angle to the spider controller script.

Step-3: Click Unity "Play" to run APE and see the physically simulated single pose walk cycle immediately. Confirm that it is as we expected.

Step-4: Back to Inverse Dynamics posing mode and drag the spider's front legs into the second pose consisted of landed left leg and a returned to neutral right leg. Since the right leg is already in its neutral position only the left leg need to be posed.

Step-5: Read the pose out of the motor sensors on the leg joints and copy them into the spider controller script.

Step-6: Click the Unity "Play" button again to check if the walking cycle is as expected. Note that the controller script automatically create the two mirrored poses from the two poses created by users in step 1 and 4. Observe that the animation cycle of the front legs are as expected. Then unanchor the torso of the spider and observe the spider fall to the ground and crawl with only the two front legs plowing.
http://youtu.be/3GvhxElhr5M
[Sorry you have to click the above URL to play the video since a single post can only have maximum 5 embedded videos]

As we can see it only takes a couple of minutes to create a walking cycle of a pair of legs and validate it with the APE without leaving the Unity Editor. We can just repeat the same steps to pose the other 3 leg pairs easily and end up with a fully physically simulated crawling spider of 8 legs in less an hour. APE Inverse Dynamics based poser makes it fun and efficient to create such physics based character controller in WYSWYG fashion. Designers and artists will be able to polish physically real models and animations inside a live environment simulated by APE directly inside the Unity Editor.

 

Fodder for leg motion...

Note: not knocking what you've done here... this is amazing stuff. You are very close to building a truly realistic spider while simply showing off your product. Kudos!

 

Just received a question from a user:

" Hey, i'd like to ask about Articulated Physics... I've been struggling with the problem of PhysX in Unity, where if a rope gets a weight on its end, it goes crazy ( http://answers.unity3d.com/storage/temp/2505-physx.png )
Does Articulated Physics solve that problem? Also, i really haven't ever tried any imported physics ( engines? ), but how will that affect other physics based things?"

And here is a quick demo to simulate the requested scene with APE. The mass ratio between the weight and a single rope segment is 100:1. I was using a mouse spring to generate up to 2000 Newton of force in order to drag the weight or the rope around in order to show how the model reacted to external force in a physical environment.

Regarding the 2nd/3rd questions, APE will be introduced as a native plugin for performance reasons.It's directly integrated into Unity component framework and Unity editor. You will be looking into a few new APE physics components to be added to GameObjects and they will replace the standard RigidBody component that is used by PhysX. All PhysX collider types including mesh collider will be supported. APE supports more collider shapes such as true analytical cylinder/cone shapes etc. You will likely need to rebuild your physics model using APE's set of components though.

 

Can you chain lots of joints to create mechanics like this? (The arm is powered by the 3 linear motors only)

 

@Zeblote : easily can APE simulate a frontend loader like that. If you pay close attention to the video you will notice that the contact between the loader and environment feels soft and the loader is running at a little bit lower than realtime frame rate. APE has much more rigid contact and higher performance. Not only does APE supports the prismatic/slider joints and hinge joints in the loader but also ball joint, universal joint, planar joint and fixed joint which you can use to build more complex vehicles. Every single DOF of the APE joints can be optionally powered by strong motors, PD springs and restricted by joint limits.

I need some artists/modelers help to create more elaborated models/rigs to show case vehicles like that.

A.P.

 

This project looks great. I've been working on a physics-based game for a few months now, and have been mildly disappointed with the NVida PhysX system. The main thing is that the joints do not hold strongly enough, even at a pretty high "Solver Iteration Count" value (and setting it too high I've read will slow it down).

Looking at what you've accomplished so far using your Articulated Physics Engine, I think this would increase the flexibility of the game I'm working on by quite a lot. The game is centered around the user's building 'units' by piecing together 'parts', (then adding logic through a custom scripting system), and then having the system 'build' the unit and have it actually play out in a match/battle (with your being able to create, control to some extent, and destroy the units as in RTS games, as if the units were built-in). This sort of physics engine, if it worked well/didn't add a lot of side-problems, would be very valuable to the target game, because it would let the user create much more elaborate unit structures. For example, something like the front-end loader shown in the above-mentioned video.

As of now, the PhysX engine works O-K for basic units (a few blocks for the body, and four WheelCollider-based wheels), but forces the user to be pretty minimal with their design, removing a lot of the opportunity for the user to make very unique/creative ones. It's probably the greatest system-inherent (not-easily-surmountable) limitation for my game, which is why I'm becoming very interested in this project.

That said, I have some questions about the workings of this physics system, which have so far made me a little hesitant:
1) First, how 'open' will the physics system be once it's released? Will it be full-source-code access, or will it be packaged/protected so the user can't directly modify it (e.g. a DLL)?
2) How long do you plan to develop/minimally-support (i.e. bug fixes, but not necessarily new features) the engine?
3) How easily would a developer (of mainly C#) be able to make fixes to the physics engine, if they did come up, and the engine was no longer being developed? Is the C++ code behind it (or the build process) very different from standard C#, or is it complicated but still 'standard' enough in its approach/infrastructure that an intermediate developer like me would at least have a chance of solving them?

Thanks.
-Stephen

 

Could you achieve a better result for combining articulated physics and normal rigidbodies than in the video? For example when he drives into the wall some cubes fly away for no reason.

Somehow I forgot about this, it's a very important question!

 

@Stephen
That sounds an exciting RTS game centered around user generated physics models. APE sounds a great fit if you're looking into building complex mechanisms with many level of joints and challenging mass ratio (e.g. instead of setting every rigid body to mass = 1, you want to set them to varied range between 1-1000 to show case the heavy weight / impact feel of some vehicles/robots). The solver is much more stable than other commercial engines I'm aware of when it comes to such challenging situations.

Also the WheelColliders are not real physically simulated wheels. They are a few ray casts to detect the how high the car should be "hovering" in the air. You may see wheel popping and weird problems if you run wheel collider vehicle on complex terrain such as . APE is meant to simulate real wheel-ground collision and rack-on-the-pinion steering and such The monster truck demo video posted earlier in this thread is although very simplistic, fully simulated the wheel, suspension and steering with real rigid body collision, piston spring strut and steering driven by multi-joint motors.

Dynamically changing the topology of the model while the simulation is running (e.g. adding a gear to the transmission chain of a car while the car is running on the road) won't be supported in the first release. However allowing a RunTime level editor to be integrated with APE to assemble a car in the game when the car is not running is not hard to support. I'll see if I have time to add such support to the first release.

Great questions on the licensing/sales scheme. I need to think about it. By default I'm thinking to include all C# code in the package which has most of the modeling layer logics. I'll strive to make it a seamless integration with Unity and straight forward to integrate with various game projects and also offer support in the long term. The low level C++ physics solver plugin will be hard for non-physics programmer to maintain.

 

Waaaaaaaaiit.... what? Don't say this requires anything that can only be done in the unity editor

 

@Zeblote: of course. APE already supports seamless integration of discrete rigid bodies and multi-joints articulated mechanisms. The "cubes fly away" problem you mentioned in the loader video is because the "soft contact" caused interpenetration between collider shapes. Which is addressed by APE's rigid contact between standard rigid bodies and articulated mechanisms.

 

Is that good enough for you if you can build an APE model in the runtime by invoking some scripting APIs before you invoke another API to kick start the simulation of the newly built APE model ? If so, it's not hard to add as I said and I'll try to add it if it is what people need.

 

Sure that's fine, just need a way to assemble stuff in a level editor at runtime. I will have to create my own user interface for it either way, so it doesn't have to be fancy by default just needs an API to actually make it work. There also won't be any physics simulation while editing, starting it manually after loading the level is no problem.

 

It's been 3 weeks, did something cool happen yet?

 

Physx 3.3 also has an articulated Joint type, but it is under active research in the Physx SDK.. I think Unity would not implement it even in Unity 5.0, WE NEED IT..!
http://physxinfo.com/wiki/PhysX_SDK_3.x
https://developer.nvidia.com/sites/default/files/akamai/physx/Manual/Advanced.html

 

That's the first thing I looked at when I surveyed existing articulated joints support. Unfortunately it only supports ball joint and is currently in research state and there was no examples showing that it worked at all. APE supports ball, universal, hinge, prismatic and planar so far.

 

Just a quick update: I'm currently fully occupied building something on top of APE but beyond APE. It will ultimately drive a new genre of physics games being made. Unfortunately that means I don't have the time I wish I had to release APE. The situation may change when we get more resource on the project down the road.

 

Was looking forward to release of ape, but if waiting means a much better system than its worth the wait. Will the new system that's been built on top of ape be release as a seperate asset?

 

Hi @ArticulatedPhysics
I am new to unity I am working on a project like Motorized Arm. I have gone through the tutorials based on Physics of Unity but I am unable to figure out how you are achieving the same in your video. Can you please help me on how to proceed in this. Please see the arm structure in the attachment

 

Did you read the topic? He wrote a custom physics engine and it isn't released yet.

 

Ya I get that. That's why I am asking how to proceed on with the present physics engine. If you can help with it please reply

 

You can model a multi-hinge arms like in the video pretty easily using the HingeJoint with the default Physics. Problem with the default PhysX engine is that when the levels of the arm goes above 2-3 and the mass ratio between joined parts goes above 1:5 the motor will become very week and the hinges start to wobble around. APE fixes such problems with a new constraint solver optimized for articulate mechanisms.

 

Nice job! I'm pretty sure my company is interested in APE.

 

@ArticulatedPhysics APE Looks very promising. I'd like to know if you have a multithreading implementation and if it works with standard unity mesh colliders.
I have to implement a proper wheel collider and suspensions for our hill climbing game and from your videos looks like that it's exactly what I need. I'm also testing Unity 5 and we are working on Android now. Please consider the option to give us access to an alpha version so that I will be able to test it and decide if to try different solutions.
Thanks

 

@mcmorry: It's currently single threaded, part of the algorithm such as broad/narrow phase collision might be easy to decouple into different thread, but I have no plan working on it yet. The collision code is SIMD accelerated with iOS / apple library. And Yes the standard mesh collider is supported by APE. It supports only convex mesh for dynamic rigid body and both convex and concave for static mesh. Of course you can run a convex decomposition algorithm (not part of APE yet, it's not hard to do in C# anyway) to decompose a concave mesh into multiple convex ones and message them together as a "compound shape" that is now supported by APE. I'm now occupied by a critical project involving writing a higher level simulation engine on top of APE. I'll see what I can do on a possible release of an early version of APE since the demands are climbing up but I can't promise anything yet. Thank you for your interest.

 

AP, how well does this thing simulate tank track?

 

Of course. It will do better than this:

 

Haha love the little dude jumping out at the end.
I really want this asset, Any eta on release, would make a nice Christmas gift to myself

 

This is absolutely amazing. As one who loves highly realistic simulation games, I was rather startled by the limitations of PhysX, particularly those cursed joints, and I am delighted at what you are doing.
A question, since I am working on an RTS with lots of tanks: how does this do performance wise? I got a special simulator from the asset store to save time, but found that I couldn't have more than two tanks without an insane frame rate drop ( funny thing was, the treads were actually visuals, with practically no physics, and yet performance suffered.) What I'm working on may have several dozen tanks at a time, by the way.
I like what I saw in that tank video, especially the little driver hopping out (abandon tank!) and I'm really excited.
Can't wait! Lots of pointlessly suppressed enthusiasm here!

 

Without doing an actual test, I would guess something around 10 tanks can probably work Ok. It really depends on how complex you build the models and how many joints you spend in those wheels and tracks. You could simulate only those thanks in field of interest and simplify the far away tanks as a single rigid body for example. Another way to scale up is to do distributed physics.

http://gamasutra.com/view/news/174061/Distributed_physics_the_key_to_online_game_simulation.php

[QUOTE="IronDuke, post: 1880286, member: 626616"
A question, since I am working on an RTS with lots of tanks: how does this do performance wise? I got a special simulator from the asset store to save time, but found that I couldn't have more than two tanks without an insane frame rate drop ( funny thing was, the treads were actually visuals, with practically no physics, and yet performance suffered.) What I'm working on may have several dozen tanks at a time, by the way.
I like what I saw in that tank video, especially the little driver hopping out (abandon tank!) and I'm really excited.
Can't wait! Lots of pointlessly suppressed enthusiasm here![/QUOTE]

 

Just a quick glimpse into what I'm hard working on: It is ACE, Articulated Character Engine, that is built on top of the high performance APE to provide robust character skeleton gait simulation and advanced balance control.

The above video is captured live with all bones, joints and joint motors of the bird dynamically simulated with ACE/APE. Towards the end of the video I applied external forces via the mouse springs to disturb the bird's walking controller. As we can see it's pretty robust against various terrain, collidable objects, turning directions and external forces.

It's still at the very early stage. But it opens up possibilities building new genres of simulation based games.

We are currently looking for an experienced Game Designer on this project to make an epic action/sim/strategy game leveraging the advanced ACE/APE simulation technologies and Unity. If you are interested please drop me a PM.

Cheers,
AP

 

APE is looking awesome. I'd rather see an earlier APE release with core functionality than a later one with more advanced features/addons such as ACE.

 

I agree with SunnySunshine. APE would be VERY useful on my current project. I could wait for ACE.

 

@ArticulatedPhysics please consider to prioritize APE. It has a much bigger potential in term of use cases then ACE.

 

That's cool no doubt, but why are you diving into the next project before finishing the first? You should release a general purpose APE first!

 

Did something happen? Don't leave us without relasing APE!

 

This looks awesome! I have been working on creating arrays of Unity joints through scripts connecting cubes together basically for destruction purposes. Things just do not behave as I would have expected. Creating the joints in script is a pain as the joint anchor positions apparently get created where the joint is connected and cant be re-positioned easily. How many joints do you think this system could handle? Multiple joints per object also?

By the way, I created this spider animation using only IK some time ago using another engine. It simply moves the legs back to a nominal position once they move outside a certain radius from nominal. When you move the body around using physics forces the legs actually move by following the body as though they are walking. Makes it look like the legs are doing the pushing. Would be nice to not have to fake it

So I also want to recreate this with APE/ACE being able to not only move the spider with motors but also rip the limbs off one by one LOL. Yes I'm sadistic that way.

 

Right OP didn't even visit the forums since over 3 weeks ago so I guess we can forget this.

Which sucks as there is still no alternative.

 

I wouldn't give up quite yet. He may be on a frantic sprint to get a releasable version out, or he might be caught up in post-holiday stuff. That's what has kept me from posting for a while.
I sure hope he hurries, though. My RTS game is completely frozen till APE comes out, because the whole thing is centered on tanks and other vehicles.
@ArticulatedPhysics , if you see this... Could APE possibly be released without further improvement? Like a quick first version with updates to follow? Because what I saw in that tank video is absolutely perfect for my needs, and I could improve on it if needed. I'd really like it now, but if it can't be done, that's OK. I can probably wait a couple more weeks, and work on the pinball game I started to pass the time. I am however leaning somewhat toward getting the Physics Tank Maker from the Asset Store instead.

--IronDuke

 

The IK spider looks really cool. If you don't need much interaction with the environment physically IK driven animation fits perfectly the need. Physical simulation of a spider only justifies itself if you slam the spider with a rock or tie its legs with ropes. It would be very challenging but doable.

 

Sorry for not finding the time to release APE. My own game production took all my time and we have a clock ticking. My current priority is to hit the 1st playable milestone of our game in the next few months while completing APE/ACE. The engine is definitely not abandoned. Not only are we continuing to work on it but also we are investing a lot of time to build an asset pipeline around it so that physics assets can be authored in Maya and simulated/tuned in Unity seamlessly. Because once you have a powerful engine you will spend tons of time to do modeling and physics parameters optimization which will become the bottleneck to build a deep simulated game -- I know it because that's what we're experiencing

I will re-visit the possibility of releasing something when we finish our 1st playable. But I can't promise anything. So please don't bet on it.

As always thank you for the interest and support.

-ap

 

And he's alive! Yeah! I was actually getting worried.
So you are not only making the bare framework of APE, but actually getting stuff in for maximum use of its features AND making your own game?!
Wow... I definitely see now why it's taking so long. Keep us updated on the progress of your game as well as APE, because I have a funny feeling it will be a pretty good one.
Adding functionality to save time when using APE is very smart. Time is the least easily renewable resource, after all. I hadn't thought of doing that, as I'm not the best programmer in the world...
Last, sorry if I seem to have pressed you. I was rather impatient in the last post, but I love everything you have done so far. Keep going at your own pace, by all means!
--IronDuke

 

hii x i just want to register my interest.

Right now i'm hitting a wall: http://forum.unity3d.com/threads/fixed-joint-forces-reporting.298782/
I think physX is pretty fine for my needs if it worked as documented, but i'm being held up because of a crippling bug.

What i basically need to do is setup a wall made of precalculated chunks (can use cubes as a simulation stand in) and connect all the chunks that touch each other with joints, which break under a certain stress force. I also need some way to detect how much force is put on a chunk when it's hit hard, but not hard enough to break it off, so that i can do some scripted incremental damage things to it (weakening its joint connections, applying crack decals, etc).

could your system handle this? I'm guessing its pretty elementary by comparison

 

Contact forces at any contact point of a contact manifold can be easily read out by an API in APE. For "max force" to break a joint, if you configured 100N maximum force for the joint, isn't it always 100N or close when it is broken and why do you need to read out the breaking force ? Or are you talking about a sensor to measure the dynamic stress force at the joint before the joint is breaking ? I *think* it's not hard to provide an API to read out the instaneuous joint force/impulse to hold the joint together, it will be highly varying though.

Kevin

 

My situation looks like this:



Each thin red line is a joint, each node is a chunk. It's a wall built out of pre-fractured chunks (they're all individual objects)

When enough force is applied to one chunk, i'd like the joints on it to break, thus seperating it from the wall. When a lesser force is applied, i'll have the joints degrade (by adjusting their strength) and the chunk itself will also be damaged, with crack decals and such.

So i need to know, roughly, the stress on the joints, or on a chunk.

 

Hi, ArticulatedPhysics

I'm working on robotics simulation for testing prototypes and I think APE is exactly what we are looking for. We simulate robotics arms, cables (attached to bodies with big mass) and vehicles. They are bascically rigid bodies linked by (a lot of) joints and motors. Usually, their mass values are very different (small bodies attatched to big bodies).
At first, we tried to use PhysX, but , for the reasons you've pointed, it didn't work. Right now, we are using a dll for accessing ODE engine in Unity, but performance and stability aren't nice enough. Cables keep exploding when they are very long, for example.

Your demonstations really caught our attention. It seems ideal for us. We would be very interested in testing APE. Any date for release? Is there a possibility of testing it before releasing?

Keep the good work,

Eduardo de Albuquerque
Tecgraf PUC-RIO
http://www.tecgraf.puc-rio.br/

 

Hello AP,

Yes, excellent implementation for robotics work. What's your progress for ACE/APE release?

Regards,

--RW

 

Wow, this looks great. I'd also love to know when, approximately, this might be released.

 

This is exactly what I need for my car simulator project, hope it will be released soon !

 

My bets are on "never" by now. This is like spectra gi.

 

A tech demo of a fully physics simulated fat chicken action game, Fat Chicken Fighting, powered by APE, the advanced physics engine for robust joint and skeleton motion