Agent won't move after upgrade from 0.13 to 0.15

I decided to upgrade my project because 0.15 is a release candidate, but I think my project broke. I don't use deprecated methods, I also adopted change from 0.14 but for some reason my agent won't move be it Heuristic or Inference (sample project works fine), cpu usage is also high. Their was an error in the editor that disappeared after I restarted unity. I don't have the full error message but I manage to salvage some, it said, "Could not load signature of MLAgents.Demonstration Recorder:Initialize DemoStore... Could not load file or assembly 'System.IO.Abstractions..."

What I've done so far

• InitializeAgent() to Initialize()
• AgentAction() to OnActionReceived()
• AgentReset() to OnEpisodeBegin()
• Done() to EndEpisode()
  
• Create a new MonoBehaviour and store the data in the new MonoBehaviour instead of Inhering from Academy
  
• Move the InitializeAcademy code to MonoBehaviour.OnAwake
• Use Barracuda 0.6.2, 0.6.1, 0.6.0 and 0.5.0
• Check Academy.k_ApiVersion ("0.15.0")

What I failed to do

• Check UnityEnvironment.API_VERSION in environment.py (I can't find it's location)

 

DecisionRequester was the answer. Thank you so much.

 

Decision requester is a band-aid component to help people switch to the new system where RequestDecision() is a requirement.

On demand decision making is more difficult to get right when you are making decisions are variable intervals, but it is also much more efficient, and much more interesting from a generalized learning perspective.

In any case, I recommend that people create simple counters in their agent classes which can track fixed update ticks and request decisions according to a set interval. Here's an example:

 

Code (CSharp):

1. public int maxSteps = 512;
2.         public int decisionSparsity = 10;
3.         public int currentStep;
4.         public int decisionCounter = 0;
5.      
6.         public void RunDecisionCounter()
7.         {
8.                 if (currentStep > maxSteps)
9.                 {
10.                     EndEpisode();
11.                     decisionCounter = decisionSparsity;
12.                 }
13.                 else
14.                 if (decisionCounter <= 0)
15.                 {
16.                     RequestDecision();
17.                     decisionCounter = decisionSparsity;
18.                 }
19.                 decisionCounter -= 1; // decrements decision counter every fixed update
20.         }
21.      
22.         public override void OnEpisodeBegin() // formerly agent reset
23.         {
24.                 currentStep = 0;
25.         }
26.         public override void OnActionReceived(float[] vectorAction)
27.      
28.                 AddReward(points);
29.                 // agent action stuff here
30.                 currentStep++;
31.         }
32.         public void FixedUpdate()
33.         {
34.                 RunDecisionCounter();
35.         }