AI Planner - v0.1.1-preview Release!

Can't wait to explore your amazing tool!!
The problem is that I can't run otto. I use Github Desktop to clone the project, and has a 26KB Domain.dll. And I use mac 10.14 & Unity 2019.3.04b. Wondering anything I did wrong?

Spoiler

InvalidOperationException: The native container has been declared as [ReadOnly] in the job, but you are writing to it.
Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle.CheckWriteAndThrowNoEarlyOut (Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle handle) <0x1951026d0 + 0x00052> in <f55fb603e84f4928ae3dd5f6d0b3e174>:0
Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle.CheckWriteAndThrow (Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle handle) (at /Users/builduser/buildslave/unity/build/Runtime/Export/Jobs/AtomicSafetyHandle.bindings.cs:153)
Unity.Entities.EntityCommandBuffer+Concurrent.CheckWriteAccess () (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityCommandBuffer.cs:1562)
Unity.Entities.EntityCommandBuffer+Concurrent.Instantiate (System.Int32 jobIndex, Unity.Entities.Entity e) (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityCommandBuffer.cs:1598)
AI.Planner.Domains.StateData.Copy (System.Int32 jobIndex, Unity.Entities.EntityCommandBuffer+Concurrent entityCommandBuffer) (at <28519cad2e124d71a99d69d55e3fb90e>:0)
AI.Planner.Domains.StateDataContext.CopyStateData (AI.Planner.Domains.StateData stateData) (at <28519cad2e124d71a99d69d55e3fb90e>:0)
AI.Planner.Actions.WorkaholicAgent.Navigate.ApplyEffects (Unity.AI.Planner.DomainLanguage.TraitBased.ActionKey action, AI.Planner.Domains.StateEntityKey originalStateEntityKey) (at <7e7d171223244862a70018811e407552>:0)
AI.Planner.Actions.WorkaholicAgent.Navigate.Execute (System.Int32 jobIndex) (at <7e7d171223244862a70018811e407552>:0)
Unity.Jobs.IJobParallelForDeferExtensions+JobStructDefer`1[T].Execute (T& self, System.IntPtr additionalPtr, System.IntPtr bufferRangePatchData, Unity.Jobs.LowLevel.Unsafe.JobRanges& ranges, System.Int32 jobIndex) (at Library/PackageCache/com.unity.jobs@0.1.1-preview/Unity.Jobs/IJobParallelForDefer.cs:105)

 

Yes I know that, I guess Github desktop automatically handle it. The otto folder is 356.1M after Clone. Then I open it in Unity, it becomes 896M. (Is it correct? So it's not lfs problem right?)

 

I use 2019.2 and run it correctly, such a surprise! The bug report sounds it's nothing related to the Unity version. Thanks for your help!

 

Happy to hear an update about this project, but I am still confused as to why the GameObject workflow is still being used. Is there a firm ETA on when this package will be usable in a pure ECS context?

 

DOTS GameObject to Entity workflows are still in the process of being created, so DOTS usage of the planner would primarily be code-based currently. And currently, the planner should be able to be used code-based-wise. Do you have a DOTS-based project you are already wanting to use the planner on?

 

Do you envision timing a pure ECS version when the core API hits 1.0, or do you think the authoring workflow won't be in a good enough spot at that point still?

And yea I have a project booting up in a few months that would be a good candidate for this system. I've implemented a pure-ECS utility AI system for my current project, but from a workflow point of view this project is much more interesting to me.

 

I can't comment on DOTS release plans, but once the authoring workflow has stabilized we will support that as best as we can.

However, that brings me back to what you would be looking for re: your DOTS/ECS project. Are you planning to use a visual authoring workflow for that project or will you be working mostly from code?

 

What @growling_egg says.

And also screen capture one of your guys making a typical GOAP agent like a farmer (harvest vegetables, when tool broken gather parts and uses anvil to fix a new tool etc...), with sphere capsules and cubes. Even unedited that'll give us a thinking process to follow.

Visual.

Or download the zipped project from : https://drive.google.com/open?id=1BXLMI6NZFEAhqIxsj_2qqiW8vx7_A6nu
If you want Github to fix their website LFS zip download, all of you contact them from this page https://support.github.com/contact
In the meantime, Unity you can zip the Otto project and add it to release, this way it can be deployed by simply downloading it

 

Actually this reminds me to talk about my project. It's a pure ECS project too. The gameplay is multi-villager management similar to Rimworld or Oxygen Not Included.

I am currently stuck at AI, since every facility in the game requires villagers to build and operate. I need villagers' behaviors to glue everything up to run the village.
I have researched three types of AI ( FSM, Behaviour Tree, GOAP) and finally decided to use GOAP which I think is the fittest to my game. The problem is there's no out-of-box GOAP addon working for ECS World out there. Then I found your AI Planning project, Similar to GOAP and developing in Unity itself, and is ECS cored inside! Which is perfect to me!
I apologize for so many non-sense talking not relevant to AI Planner, and my bad English. I am really anxious about my working progress, because it's been developed for 2 years, and I promised to release the demo this year. So I don't need visual authoring workflow that much, I would like to be working mostly from code, as long as I can implement AI and continue developing as soon as possible.

 

Do you know when you'll support 2019.3?

[Edit]Also where is the best place to submit bugs/issues and help out? it's a very interesting project!

I'm getting these errors in b5

Code (csharp):

1. InvalidOperationException: The native container has been declared as [ReadOnly] in the job, but you are writing to it.
2. Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle.CheckWriteAndThrowNoEarlyOut (Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle handle) (at <5e35e4589c1948aa8af5b8e64eea8798>:0)
3. Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle.CheckWriteAndThrow (Unity.Collections.LowLevel.Unsafe.AtomicSafetyHandle handle) (at <5e35e4589c1948aa8af5b8e64eea8798>:0)
4. Unity.Entities.EntityCommandBuffer+Concurrent.CheckWriteAccess () (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityCommandBuffer.cs:1562)
5. Unity.Entities.EntityCommandBuffer+Concurrent.AddBuffer[T] (System.Int32 jobIndex, Unity.Entities.Entity e) (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityCommandBuffer.cs:1635)
6. AI.Planner.Actions.Worker.Navigate.Execute (System.Int32 jobIndex) (at <ff180c3d585443c1998ba955ec13f4e2>:0)
7. Unity.Jobs.IJobParallelForDeferExtensions+JobStructDefer`1[T].Execute (T& self, System.IntPtr additionalPtr, System.IntPtr bufferRangePatchData, Unity.Jobs.LowLevel.Unsafe.JobRanges& ranges, System.Int32 jobIndex) (at Library/PackageCache/com.unity.jobs@0.1.1-preview/Unity.Jobs/IJobParallelForDefer.cs:105)
8. Unity.Jobs.JobHandle:ScheduleBatchedJobsAndCompleteAll(Void*, Int32)
9. Unity.Jobs.JobHandle:CompleteAll(NativeArray`1)
10. AI.Planner.Actions.Worker.ActionScheduler:Schedule(JobHandle)
11. Unity.AI.Planner.PlannerScheduler`8:ScheduleActionsJobs(JobHandle) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Planner/PlannerScheduler.cs:247)
12. Unity.AI.Planner.PlannerScheduler`8:Schedule(JobHandle) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Planner/PlannerScheduler.cs:104)
13. UnityEngine.AI.Planner.DomainLanguage.TraitBased.BaseAgent`10:Update()
14. WorkerAgent:Update() (at Assets/Code/Ai/Planner/WorkerAgent.cs:31)

and

Code (CSharp):

1. ArgumentException: A component with type:NavigateFixupReference [B] has not been added to the entity.
2. Unity.Entities.EntityComponentStore.AssertEntityHasComponent (Unity.Entities.Entity entity, Unity.Entities.ComponentType componentType) (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityComponentStoreDebug.cs:216)
3. Unity.Entities.EntityComponentStore.AssertEntityHasComponent (Unity.Entities.Entity entity, System.Int32 componentType) (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityComponentStoreDebug.cs:222)
4. Unity.Entities.EntityManager.GetBuffer[T] (Unity.Entities.Entity entity) (at Library/PackageCache/com.unity.entities@0.1.1-preview/Unity.Entities/EntityManagerAccessComponentData.cs:315)
5. AI.Planner.Actions.Worker.ActionScheduler.Schedule (Unity.Jobs.JobHandle inputDeps) (at <ff180c3d585443c1998ba955ec13f4e2>:0)
6. Unity.AI.Planner.PlannerScheduler`8[TStateKey,TActionKey,TStateManager,TStateData,TStateDataContext,TActionScheduler,THeuristic,TTerminationEvaluator].ScheduleActionsJobs (Unity.Jobs.JobHandle inputDeps) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Planner/PlannerScheduler.cs:247)
7. Unity.AI.Planner.PlannerScheduler`8[TStateKey,TActionKey,TStateManager,TStateData,TStateDataContext,TActionScheduler,THeuristic,TTerminationEvaluator].Schedule (Unity.Jobs.JobHandle inputDeps) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Planner/PlannerScheduler.cs:104)
8. UnityEngine.AI.Planner.DomainLanguage.TraitBased.BaseAgent`10[TAgent,TObject,TStateKey,TStateData,TStateDataContext,TActionScheduler,THeuristic,TTerminationEvaluator,TStateManager,TActionKey].Update () (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Agent/BaseAgent.cs:275)
9. WorkerAgent.Update () (at Assets/Code/Ai/Planner/WorkerAgent.cs:31)

 

For our next release we plan to support 2019.3 as the minimum version.
For submitting bugs, please post on the forum thread for each release version. As we move closer to release we'll have a longer term plan for bug submissions either via GitHub issues or something similar.

Re: the errors. Did you make any local changes to your project. Are you using a different version of of the Unity.Entities package?

 

That's awesome!

I'm testing on a real project so there are loads of changes. I'll try starting a bare bones project and see how it goes.
Unity.Entities is updated(0.1.1-preview). Here's my Packages manifest:

Spoiler: manifest.json

"dependencies": {
"com.unity.2d.sprite": "1.0.0",
"com.unity.2d.tilemap": "1.0.0",
"com.unity.addressables": "1.1.10",
"com.unity.ads": "2.0.8",
"com.unity.ai.planner": "0.1.1-preview",
"com.unity.analytics": "3.3.2",
"com.unity.burst": "1.1.2",
"com.unity.cloud.userreporting": "0.2.4-preview",
"com.unity.collections": "0.1.1-preview",
"com.unity.entities": "0.1.1-preview",
"com.unity.ext.nunit": "1.0.0",
"com.unity.ide.rider": "1.1.1",
"com.unity.ide.visualstudio": "1.0.11",
"com.unity.ide.vscode": "1.1.2",
"com.unity.jobs": "0.1.1-preview",
"com.unity.mathematics": "1.1.0",
"com.unity.memoryprofiler": "0.1.0-preview.7",
"com.unity.performance.profile-analyzer": "0.4.0-preview.6",
"com.unity.polybrush": "1.0.1",
"com.unity.purchasing": "2.0.6",
"com.unity.rendering.hybrid": "0.1.1-preview",
"com.unity.test-framework": "1.1.3",
"com.unity.textmeshpro": "2.0.1",
"com.unity.timeline": "1.2.2",
"com.unity.ugui": "1.0.0",
"com.unity.modules.ai": "1.0.0",
"com.unity.modules.androidjni": "1.0.0",
"com.unity.modules.animation": "1.0.0",
"com.unity.modules.assetbundle": "1.0.0",
"com.unity.modules.audio": "1.0.0",
"com.unity.modules.cloth": "1.0.0",
"com.unity.modules.director": "1.0.0",
"com.unity.modules.imageconversion": "1.0.0",
"com.unity.modules.imgui": "1.0.0",
"com.unity.modules.jsonserialize": "1.0.0",
"com.unity.modules.particlesystem": "1.0.0",
"com.unity.modules.physics": "1.0.0",
"com.unity.modules.physics2d": "1.0.0",
"com.unity.modules.screencapture": "1.0.0",
"com.unity.modules.terrain": "1.0.0",
"com.unity.modules.terrainphysics": "1.0.0",
"com.unity.modules.tilemap": "1.0.0",
"com.unity.modules.ui": "1.0.0",
"com.unity.modules.uielements": "1.0.0",
"com.unity.modules.umbra": "1.0.0",
"com.unity.modules.unityanalytics": "1.0.0",
"com.unity.modules.unitywebrequest": "1.0.0",
"com.unity.modules.unitywebrequestassetbundle": "1.0.0",
"com.unity.modules.unitywebrequestaudio": "1.0.0",
"com.unity.modules.unitywebrequesttexture": "1.0.0",
"com.unity.modules.unitywebrequestwww": "1.0.0",
"com.unity.modules.vehicles": "1.0.0",
"com.unity.modules.video": "1.0.0",
"com.unity.modules.vr": "1.0.0",
"com.unity.modules.wind": "1.0.0"
}

 

Hmm...Looking at your manifest I thought I might find a newer version of one of the dependent packages in there, but they all seem to be correct. Let me know how the bare bones project goes.

 

Thanks for sharing this! Had a bit of fun playing around with Otto. I found that by just moving the stations a bit further away, the AI simulation kind of falls apart very quickly.



Sometimes Otto will loop endlessly at a workstation, sometimes he'll recover after a few loops and use the station (adjust colliders maybe?). I've also found that Otto sometimes just sleeps over and over repeatedly while he dies from hunger and/or thirst (while he still has bottles/apples in inventory).

 

Hmm...That's odd because Otto's custom navigate effects and reward are based on distances between locations.

I took a look at your scene and this is an operational model issue. You are correct in that the colliders should be adjusted because Otto keeps doing 360 rotations when he reaches a location because the collider wasn't hit. This affects the plan's expectation of how long it should take to do something like consume a bottle. I think because these time estimations grow more and more apart that Otto eventually dies.

Separate from that is how we modeled the needs and how quickly they escalate. It's possible that it isn't possible to keep Otto alive with the longer distances that have to be traveled across. If you look at the precondition for Consume, Otto will only consume when the urgency is greater than the need reduction (currently set when it is at 50%).

All-in-all, you've certainly exposed the need for better domain modeling as this domain model is closely tied to the design of the level.

 

If AI Planner tutorials are still being discussed, and if domain modeling is something that we as game developers have to become skilled at, I might suggest adding this to your tutorials/docs as something to keep in mind. Certainly it would be nice to be able to have dynamic level elements which result in dynamic kinds of plans and situations. I'd be interested to know what kinds of practices are good and bad for domain modeling.

 

That's a tad too restrictive for us. Any reason that makes our life better?

 

Hi, I configure a small scene mostly like Otto. I think I got the main idea, but it just can't run. What's the problem?

Spoiler

NullReferenceException: Object reference not set to an instance of an object
UnityEngine.AI.Planner.DomainLanguage.TraitBased.PlanningDomainData`5+<>c[TObject,TStateKey,TStateData,TStateDataContext,TStateManager].<GetInitialState>b__12_1 (UnityEngine.AI.Planner.DomainLanguage.TraitBased.TraitObjectData t) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/TraitBasedLanguage/PlanningDomainData.cs:82)
System.Linq.Enumerable+SelectListIterator`2[TSource,TResult].ToArray () (at <fbb5ed17eb6e46c680000f8910ebb50c>:0)
System.Linq.Enumerable.ToArray[TSource] (System.Collections.Generic.IEnumerable`1[T] source) (at <fbb5ed17eb6e46c680000f8910ebb50c>:0)
UnityEngine.AI.Planner.DomainLanguage.TraitBased.PlanningDomainData`5[TObject,TStateKey,TStateData,TStateDataContext,TStateManager].GetInitialState (Unity.Entities.EntityManager entityManager, System.Collections.Generic.List`1[T] initialStateData) (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/TraitBased

Spoiler

NullReferenceException: Object reference not set to an instance of an object
UnityEngine.AI.Planner.DomainLanguage.TraitBased.BaseAgent`10[TAgent,TObject,TStateKey,TStateData,TStateDataContext,TActionScheduler,THeuristic,TTerminationEvaluator,TStateManager,TActionKey].Update () (at Library/PackageCache/com.unity.ai.planner@0.1.1-preview/Runtime/Agent/BaseAgent.cs:273)
Mu.Update () (at Assets/Scripts/Mu.cs:31)

 

I have noticed that if you move the stations during the simulation Otto simply goes to it's original location.

 

The thing is that he's at 95% or so in rest ... (since he just got up from resting on bed) and even though hunger and thirst are both at below 20%, he choose to just sleep again instead of consuming the bottles/apples he has in his inventory.

 

OK I didn't know the planner is ECS. ECS makes life easier. I'll switch to 2019.3.
And wow! multithreaded and burst compiled planner?! I can totally use that for my village. Do you you cut up the planning per agent?

And tell me, it must be a challenge for you to do a planner cleanly. What I mean is GOAP usually creates strings to keep track of chains of actions during the planing process, strings I've seen produce a lot of allocations. How have you gone around those allocations?

(could you put the zipped url for otto in the original post? I reinstalled my machine and when re setting unity I noticed this example project isn't in zip in the description of the post, others might want to bite in it)

 

And speaking of Otto, I'm getting this in 2019.3b7

 

I can't get AI Planner working in 2019.3.0b6 either.

Note: Yes I read the documentation. Was testing 2019.3.0b6 for science.

 

Very interesting, thank you.
And I didn't read all the doc so I didn't know it was generating code, the performance makes more sense now.

 

Got it, is there a tool to visualize what the agent is thinking?

 

Here's where you find it.

 

Got it thanks.

Started reading the code, it's quite verbose for such a simple agent. Any plan to make it more terse? At least more readable would be good. Also comments please coz it might seem trivial to you devs but it reads like war and peace.

also this

no red in the console so it doesn't seem to be slow by error spam.

 

do you want everyone to come on board? then all of it
After all, when you create a new script Update() and Start() get their comment.

ex:
before all declarations because unity has a new fashion in c#. It used to be simple stuff like void Update(), then it became the fashion of callbacks and now it's the fashion of whatever this stuff is called

Code (CSharp):

1. public class Otto : BaseAgent<Otto, DomainObject, StateEntityKey, StateData, StateDataContext, ActionScheduler, NeedHeuristic, TerminationEvaluator, StateManager, ActionKey>

when I read this I'm like, uh what's that class doing? what are those types about? how does it work? how does it fit together and gets called? that's a lot of food to take at once, not good for digestion

and a comment before

Code (CSharp):

1. #if PLANNER_ACTIONS_GENERATED

stating where was it generated from (I think it gets generated from assets) just to get new people onboard.

and in front of all these overrides

Code (CSharp):

1.     public override void BeginExecution(StateData state, ActionKey action, Otto actor)
2.  

most of it seems to be codegen, so just add documention inline, with a "Comment All" option on the codegen panel so when we're used to all that, training wheels off

while I'm reading this code why do you use long and not unsigned int for need counter?

Code (CSharp):

1. public class ConsumeAction : OttoAction
2. {
3.     static readonly int k_Consumables = Animator.StringToHash("Consumables");
4.     static readonly int k_EatFromPocket = Animator.StringToHash("EatFromPocket");
5.     static readonly int k_DrinkFromPocket = Animator.StringToHash("DrinkFromPocket");
6.  
7.     ConsumableType m_ConsumableType;
8.     NeedType       m_NeedType;
9.     long           m_NeedReduction;

and what's that? IOperationalAction

 

The ML stuff is really interesting.
I was just wondering if Unity was doing anything with concepts of threat IE from differentiating safe from danger?
I guess you could create your own threat system but having agent behavior either in a cooperative, neutral, or adversarial capacity is cool.
Just wondered are these just example states and could be any such as defensive, neutral, or aggressive depending on what you place in a state list or are these more set in stone?

 

I may be misunderstanding your question, but I can say that the planner is general purpose. So, it's definitely possible to create your own domain model that has modes of cooperative, neutral, or adversarial behavior, which could be specified in a state. Nothing at that level is set in stone.

If I misunderstood your question can you give more context to what you were referring to?

 

Don't worry as its me being a noob and just firing badly phrased questions.

I watched one of the demo's where an agent would attack and then runaway if low on health.
Which I presume is just a hardcoded state switch.

I was just wondering with learning can it learn that more health than it is a threat, or I am outnumbered or they out gun me, from just prior training or is it all just like the above and hardcoded switches?

 

Hello, I just wondered how it's going? I feel like the big Unity announcements tend to skip over AI developments and it's one of the things I'm looking forward to the most.
I remember we had discussed possible documentation/training enhancements awhile ago, is there anything new in the works regarding that?
Also, are there upcoming changes to the Otto demo to look forward to?

 

Are you referring to an ML-Agents demo? I don't believe we've released or shown any demo for the AI Planner that has an agent that attacks.

 

I'll copy the response from the other (now locked) thread:
We've begun the process of preparing our next release this week, so we're planning for a November release (probably somewhere during the middle of the month). However, we're also dependent on the next Entities package release, v0.2, which is going to be released soon.

We'll have a new step-by-step guide for a first-time walk through. A video will be a nice-to-have, but I'm not sure we can commit to that yet.

Otto will be updated to match with the new workflow changes, but no additional functionality will be added.

 

it would be nice to have a DOTS conversion-based workflow guide (code-based would also be great)

 

Hello! I'm very excited to use the AI planner but I'm running into some issues. First, the documentation feels sparse. When I run into a problem understanding the documentation I turn to the example but Otto seems to be out of date. For example, the AI.Planner.Domains namespace used in Otto doesn't seem to exist in my project. I'm using 0.1.1. Does anyone have any "proof of concept" projects you'd be willing to share? Something like https://gamedevelopment.tutsplus.co...d-action-planning-for-a-smarter-ai--cms-20793 would be ideal. Is there a more in depth tutorial planned?

 

@Blargenflargle make sure your looking at the right documentation. google keeps sending me to ai.planner@0.0 when it should be 0.1 https://docs.unity3d.com/Packages/com.unity.ai.planner@0.1/manual/index.html . Same thing happened with dots. A new version should be coming soon?
@amirebrahimi_unity maybe you could ask the documentation team to add a version number/drop down

 

I was looking at those docs, and realized that some of my problems stemmed from not building my assemblies. I'm still having other problems. For instance, what is a THeuristic and how do I define it? Otto has the NeedHeuristic but I'm not sure what it does just looking at the code.

 

Is there any update regards new release.

 

Still waiting on an Entities package release, unfortunately.

 

Hi @Blargenflargle. The THeuristic argument is just a struct that implements IHeuristic<StateData>. There's not a lot of documentation there currently, but the basic idea is a heuristic allows the planner to estimate the value of a state (and what future states it may lead to) in the policy graph based on it's current values. This helps the planner decide which next state to explore in a future expansion. So, the NeedHeuristic for Otto looks at all the needs (i.e. hunger, thirst, and fatigue) and tracks that against what Otto has in inventory to address those needs. Ultimately, this results in a single value for the state. The higher the value the better (i.e. Otto will not die).

 

This is already in the works from our packages team. An example:
https://docs.unity3d.com/Packages/com.unity.2d.animation@3.0/manual/index.html

I'm told that there will also be a drop-down soon, too.