Question Discerete actiona always are 0

Guys, help. In Unity, I'm trying to make an ai using mlagents. The problem is that Discrete actions always return 0 (or rarely 1 after dancing with tambourines but it bug again after the 1st code change). I was busy for half a day, I did not find a solution on the forums (mb searched badly). In the inspector in branch N size always > 0.



Agent's code:

Code (CSharp):

1.  using System.Collections;
2. using System.Collections.Generic;
3. using UnityEngine;
4. using Unity.MLAgents;
5. using Unity.MLAgents.Actuators;
6. using Unity.MLAgents.Sensors;
7.  
8. public class TestAgent : Agent
9. {
10.     [SerializeField]
11.     private Transform _target;
12.     private Vector3 _oldPosition;
13.     private float _speed;
14.     private bool _jump;
15.     [SerializeField]
16.     private Rigidbody _rb;
17.     private bool _onGround;
18.     public override void OnActionReceived(ActionBuffers actions)
19.     {
20.         float moveX = actions.ContinuousActions[0];
21.         float moveZ = actions.ContinuousActions[1];
22.         _jump = actions.DiscreteActions[1] != 0;
23.         Debug.Log(_jump);
24.         transform.position += new Vector3(moveX, 0, moveZ) * Time.deltaTime * _speed;
25.     }
26.     public override void CollectObservations(VectorSensor sensor)
27.     {
28.         sensor.AddObservation(transform.localPosition);
29.         sensor.AddObservation(_target.localPosition);
30.     }
31.     public override void OnEpisodeBegin()
32.     {
33.         _speed = 3f;
34.         _target.localPosition = new Vector3(Random.Range(-10f, 10f), 3f, Random.Range(-15f, 15f));
35.     }
36.     private void Update()
37.     {
38.         var reward = 0.1f;
39.         if (transform.localPosition.y < -1f)
40.         {
41.             SetReward(-10f);
42.             transform.localPosition = Vector3.zero;
43.             EndEpisode();
44.             return;
45.         }
46.         _oldPosition = new Vector3(_oldPosition.x, 0, _oldPosition.z);
47.        
48.         var oldDistance = Vector3.Distance(_oldPosition, _target.localPosition - new Vector3(0f, _target.localPosition.y, 0f));
49.         var currentDistance = Vector3.Distance(transform.localPosition, _target.localPosition - new Vector3(0f, _target.localPosition.y, 0f));
50.         if (currentDistance <= 3f)
51.         {
52.             AddReward(reward * 2);
53.         }
54.         else
55.         {
56.             if (currentDistance > oldDistance)
57.             {
58.                 AddReward(-reward / 10);
59.             }
60.             else if (currentDistance < oldDistance)
61.             {
62.                 AddReward(reward / 10);
63.             }
64.         }
65.  
66.         if (_jump && _onGround)
67.         {
68.             if (currentDistance <= 3)
69.             {
70.                 AddReward(reward * 10);
71.             }
72.             _rb.AddForce(Vector3.up * 10, ForceMode.Impulse);
73.         }
74.         _oldPosition = transform.localPosition;
75.     }
76.     public override void Heuristic(in ActionBuffers actionsOut)
77.     {
78.         var continuousActions = actionsOut.ContinuousActions;
79.  
80.         var discreteActions = actionsOut.DiscreteActions;
81.         continuousActions[0] = Input.GetAxis("Horizontal");
82.         continuousActions[1] = Input.GetAxis("Vertical");
83.         discreteActions[0] = Input.GetAxis("Jump") == 0 ? 0 : 1;
84.     }
85.     private void OnTriggerEnter(Collider other)
86.     {
87.         if (other.CompareTag("Target"))
88.         {
89.             AddReward(50f);
90.         }
91.         EndEpisode();
92.     }
93.     private void OnCollisionEnter(Collision collision)
94.     {
95.         _onGround = true;
96.     }
97.     private void OnCollisionExit(Collision collision)
98.     {
99.         _onGround = false;
100.     }
101. }
102.  

My Agent's object params in inspector on Google Drive
https://drive.google.com/file/d/1f2jBTn1iC452LJCqGfnXisLXxhHJY_hN/view?usp=share_link

 

your branch sizes are 1 in your inspector - meaning they can only hold 1 value (0 is the first result)

 

It working bruh. Tysm man. but i tried it yesterday, and it wasn't work. Magic lol.

 

are you using discrete branch 0 or 1 for jumping? your heuristics is set to use branch 0 and your ai actions is set to use branch 1

make sure you can test your code is working in heuristics first, then the AI should be fine if it's using the same logic

 

It's working in heuristics and I use discreteActions[0] for jumps.