Update AI behavior in Snake Game for improved performance

This commit includes updates to the assets, scripts, and AI architecture of the Snake Game. The AI agent's decision-strategies have been modified to reinforce actions that move the agent closer to the reward, in order to improve game performance. Additionally, attributes relating to the positioning and the activity of environment features in the game level have been changed for better alignment with the new decision-logic. Overall, these changes are made to enhance the agent's learning and navigation in the game.
Score after 1HR = 36

Assets/ML-Agents/Timers/AI_timers.json

@@ -1 +1 @@
-{"count":1,"self":1900.9214464,"total":3646.702797,"children":{"InitializeActuators":{"count":20,"self":0.0020009,"total":0.0020009,"children":null},"InitializeSensors":{"count":20,"self":0.0025141,"total":0.0025141,"children":null},"AgentSendState":{"count":214702,"self":12.033138399999999,"total":59.1252904,"children":{"CollectObservations":{"count":1526209,"self":14.3286336,"total":14.328634099999999,"children":null},"WriteActionMask":{"count":1526209,"self":2.9685319999999997,"total":2.9685319,"children":null},"RequestDecision":{"count":1526209,"self":10.107839199999999,"total":29.7949861,"children":{"AgentInfo.ToProto":{"count":1516682,"self":7.8454624,"total":19.6871471,"children":{"GenerateSensorData":{"count":1516682,"self":11.841683999999999,"total":11.841684299999999,"children":null}}}}}}},"DecideAction":{"count":214702,"self":1617.9479552,"total":1617.9479155,"children":null},"AgentAct":{"count":214702,"self":68.6426944,"total":68.703616099999991,"children":{"AgentInfo.ToProto":{"count":3103,"self":0.0373284,"total":0.0609244,"children":{"GenerateSensorData":{"count":3103,"self":0.023596,"total":0.023596,"children":null}}}}}},"gauges":{"SnakeAi.CumulativeReward":{"count":3201,"max":356.897736,"min":-1060.9856,"runningAverage":117.872993,"value":-10.5,"weightedAverage":13.1152983}},"metadata":{"timer_format_version":"0.1.0","start_time_seconds":"1689105069","unity_version":"2023.1.3f1","command_line_arguments":"C:\\Program Files\\Unity\\Hub\\Editor\\2023.1.3f1\\Editor\\Unity.exe -projectpath C:\\Users\\felix\\SnakeGame -useHub -hubIPC -cloudEnvironment production","communication_protocol_version":"1.5.0","com.unity.ml-agents_version":"2.3.0-exp.3","scene_name":"AI","end_time_seconds":"1689108716"}}
+{"count":1,"self":1070.6020352,"total":3714.9059334,"children":{"InitializeActuators":{"count":24,"self":0.0015083,"total":0.0015083,"children":null},"InitializeSensors":{"count":24,"self":0.0015055,"total":0.0015055,"children":null},"AgentSendState":{"count":461466,"self":7.9821912,"total":43.912659399999995,"children":{"CollectObservations":{"count":3691728,"self":11.673304,"total":11.6733041,"children":null},"WriteActionMask":{"count":3691728,"self":1.7134748,"total":1.7134748,"children":null},"RequestDecision":{"count":3691728,"self":5.8726204,"total":22.5436892,"children":{"AgentInfo.ToProto":{"count":3691728,"self":5.2082856,"total":16.671068899999998,"children":{"GenerateSensorData":{"count":3691728,"self":11.462783199999999,"total":11.4627831,"children":null}}}}}}},"DecideAction":{"count":461466,"self":2493.6136704,"total":2493.6136833,"children":null},"AgentAct":{"count":461466,"self":106.6151168,"total":106.77455959999999,"children":{"AgentInfo.ToProto":{"count":20861,"self":0.0722235,"total":0.1594411,"children":{"GenerateSensorData":{"count":20861,"self":0.087217599999999992,"total":0.087217599999999992,"children":null}}}}}},"gauges":{"SnakeAi.CumulativeReward":{"count":20861,"max":281.898163,"min":-137.800354,"runningAverage":50.4565544,"value":35.8000336,"weightedAverage":68.77597}},"metadata":{"timer_format_version":"0.1.0","start_time_seconds":"1689110079","unity_version":"2023.1.3f1","command_line_arguments":"C:\\Program Files\\Unity\\Hub\\Editor\\2023.1.3f1\\Editor\\Unity.exe -projectpath C:\\Users\\felix\\SnakeGame -useHub -hubIPC -cloudEnvironment production","communication_protocol_version":"1.5.0","com.unity.ml-agents_version":"2.3.0-exp.3","scene_name":"AI","end_time_seconds":"1689113793"}}

Assets/SnakeGame/Scripts/SnakeAgent.cs

@@ -50,7 +50,9 @@ public float PreviousDistance
         {
             get => _previousDistance;
             set => _previousDistance = value;
-        }
+        } 
+
+
 
         #region Event Functions
 
@@ -107,6 +109,8 @@ private void InitializeGame() {
                 board = new Board(width, height, snakes);
             }
 
+            PreviousDistance = float.PositiveInfinity;
+
 
             // food
             for (int i = 0; i < foodCount; i++) {
@@ -139,7 +143,7 @@ public override void CollectObservations(VectorSensor sensor) {
 
 
             // Add the direction to the food (2 floats)
-            sensor.AddObservation(directionToFood);
+            sensor.AddObservation(food);
 
             // Check for immediate dangers: front, left, right relative to the snake's current direction
             var forwardPosition = snake.Position + snake.Direction;
@@ -285,6 +289,9 @@ public override void OnActionReceived(ActionBuffers actions) {
                 if (normalizedCurrentDistance <= rewardRadius) {
                     // Compute the reward based on the Normalized current distance
                     // float reward = Mathf.Log((snake.Length + PreviousDistance) / (snake.Length + currentDistance));
+
+
+
 
 
                     // Define the reward
@@ -297,6 +304,16 @@ public override void OnActionReceived(ActionBuffers actions) {
                     // AddReward(reward);
                     PreviousDistance = currentDistance;
                 }
+
+                if (currentDistance < PreviousDistance) {
+                    AddReward(0.01f);
+                    Debug.Log("Closer :" + 0.01f);
+
+                } else if (currentDistance > PreviousDistance) {
+                    AddReward(-0.01f);
+                    Debug.Log("Further :" + -0.01f);
+                }
+                PreviousDistance = currentDistance;
             }