Best configuration for touch input training?

Hi, I'm triying to do an AI that analize hand gestures for a "touch wall" and give the correct answer, like: right, left, up, down, right circle, etc.
My problem is that the brain isn't learning and I think that the code is quite well

For the inputs I made a json that store all the touch inputs that I record myself previously ... the first value always will be 0,0. The code get a random input when the desition finished
X = time
Y = x Pos
Z = y Pos

Example:

Code (CSharp):

1. {
2.     "Direction": "Left",
3.     "TimePos": {
4.         "0": "0, 0, 0",
5.         "1": "0.0139613, 0, 0",
6.         "2": "0.0262996, -0.04159969, 0.008319868",
7.         "3": "0.0398894, -0.07903878, 0.01247977",
8.         "4": "0.0528541, -0.1372779, 0.02079964",
9.         "5": "0.0660014, -0.2038375, 0.02911958",
10.         "6": "0.079783, -0.2953562, 0.03743952",
11.         "7": "0.0927471, -0.3910351, 0.04159942",
12.         "8": "0.1057109, -0.5075134, 0.04575932",
13.         "9": "0.1189348, -0.6364714, 0.04575932",
14.         "10": "0.1326411, -0.7820694, 0.04575932",
15.         "11": "0.1453997, -0.9401475, 0.04159942",
16.         "12": "0.1586328, -1.056626, 0.03743952",
17.         "13": "0.1718672, -1.156465, 0.03327948",
18.         "14": "0.1850997, -1.202224, 0.02911958",
19.         "15": "0.1984685, -1.210544, 0.02911958",
20.         "16": "0.2115677, -1.210544, 0.02495961",
21.         "17": "0.2253915, -1.210544, 0.02495961"
22.     }

The program reads the gesture values in every frame, when it finished it calls "RequestDecision();"
Code:

Code (CSharp):

1.     private void Update()
2.     {
3.    
4.         data = new Vector3 (jsonValues.timePos[indexData].x, jsonValues.timePos[indexData].y, jsonValues.timePos[indexData].z);
5.  
6.         CollectObservations();
7.         indexData++;
8.  
9.         if (indexData >= numberData)
10.         {
11.             RequestDecision();
12.             GetRandomValues();
13.         }
14.     }

The Agent Action code:

Code (CSharp):

1.     public override void AgentAction(float[] act)
2.     {
3.         string dataMove = "";
4.  
5.         switch (act[0])
6.         {
7.             case 0:
8.                 dataMove = "Right";
9.                 break;
10.             case 1:
11.                 dataMove = "Left";
12.                 break;
13.         }
14.  
15.         realText.text = jsonValues.direction;
16.         AItext.text = dataMove;
17.  
18.         if (dataMove == jsonValues.direction)
19.         {
20.             AddReward(0.1f);
21.           //  goodOrBadTxt.text = "Good";
22.            // GetRandomValues();
23.            // Done();
24.         }
25.         else
26.         {
27.             AddReward(-0.05f);
28.          //   goodOrBadTxt.text = "Bad";
29.             Done();
30.         }
31.  
32.  
33.     }

Vector Obs code:

Code (CSharp):

1.     public override void CollectObservations()
2.     {
3.         AddVectorObs(data.y);
4.         AddVectorObs(data.z);
5.         //    Debug.Log("Obs " + data + " index data " + indexData);
6.  
7.     }

Other functions;

Code (CSharp):

1. void GetRandomValues()
2.     {
3.         indexData = 0;
4.         int ran = Random.Range(0, jsonValuesArray.Length);
5.  
6.         jsonValues = jsonValuesArray[ran];
7.         numberData = jsonValues.timePos.Length;
8.     }
9.  
10.  
11.     public override void AgentReset()
12.     {
13.         SetResetParameters();
14.     }
15.  
16.     public void SetResetParameters()
17.     {
18.         indexData = 0;
19.     }

Also I enabled the "On Demand Decisions" checkbox

Any idea?
Greetings

 

My bad, I change Update to FixedUpdate and now is learning