Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 

17 KiB

Making a New Learning Environment

This tutorial walks through the process of creating a Unity Environment from scratch. We recommend first reading the Getting Started guide to understand the concepts presented here first in an already-built environment.

A simple ML-Agents environment

In this example, we will create an agent capable of controlling a ball on a platform. We will then train the agent to roll the ball toward the cube while avoiding falling off the platform.

Overview

Using the ML-Agents Toolkit in a Unity project involves the following basic steps:

  1. Create an environment for your agents to live in. An environment can range from a simple physical simulation containing a few objects to an entire game or ecosystem.
  2. Implement your Agent subclasses. An Agent subclass defines the code an Agent uses to observe its environment, to carry out assigned actions, and to calculate the rewards used for reinforcement training. You can also implement optional methods to reset the Agent when it has finished or failed its task.
  3. Add your Agent subclasses to appropriate GameObjects, typically, the object in the scene that represents the Agent in the simulation.

Note: If you are unfamiliar with Unity, refer to Learning the interface in the Unity Manual if an Editor task isn't explained sufficiently in this tutorial.

If you haven't already, follow the installation instructions.

Set Up the Unity Project

The first task to accomplish is simply creating a new Unity project and importing the ML-Agents assets into it:

  1. Launch Unity Hub and create a new 3D project named "RollerBall".
  2. Add the ML-Agents Unity package to your project.

Your Unity Project window should contain the following assets:

Project window

Create the Environment

Next, we will create a very simple scene to act as our learning environment. The "physical" components of the environment include a Plane to act as the floor for the Agent to move around on, a Cube to act as the goal or target for the agent to seek, and a Sphere to represent the Agent itself.

Create the Floor Plane

  1. Right click in Hierarchy window, select 3D Object > Plane.
  2. Name the GameObject "Floor."
  3. Select the Floor Plane to view its properties in the Inspector window.
  4. Set Transform to Position = (0, 0, 0), Rotation = (0, 0, 0), Scale = (1, 1, 1).

The Floor in the Inspector window

Add the Target Cube

  1. Right click in Hierarchy window, select 3D Object > Cube.
  2. Name the GameObject "Target"
  3. Select the Target Cube to view its properties in the Inspector window.
  4. Set Transform to Position = 3, 0.5, 3), Rotation = (0, 0, 0), Scale = (1, 1, 1).

The Target Cube in the Inspector window

Add the Agent Sphere

  1. Right click in Hierarchy window, select 3D Object > Sphere.
  2. Name the GameObject "RollerAgent"
  3. Select the RollerAgent Sphere to view its properties in the Inspector window.
  4. Set Transform to Position = (0, 0.5, 0), Rotation = (0, 0, 0), Scale = (1, 1, 1).
  5. Click Add Component.
  6. Add the Rigidbody component to the Sphere.

The Agent GameObject in the Inspector window

Note that the screenshot above includes the Roller Agent script, which we will create in the next section. However, before we do that, we'll first group the floor, target and agent under a single, empty, GameObject. This will simplify some of our subsequent steps.

The Hierarchy window

To do so:

  1. Right-click on your Project Hierarchy and create a new empty GameObject. Name it TrainingArea.
  2. Reset the TrainingArea’s Transform so that it is at (0,0,0) with Rotation (0,0,0) and Scale (1,1,1).
  3. Drag the Floor, Target, and RollerAgent GameObjects in the Hierarchy into the TrainingArea GameObject.

Implement an Agent

To create the Agent:

  1. Select the RollerAgent GameObject to view it in the Inspector window.
  2. Click Add Component.
  3. Click New Script in the list of components (at the bottom).
  4. Name the script "RollerAgent".
  5. Click Create and Add.

Then, edit the new RollerAgent script:

  1. In the Unity Project window, double-click the RollerAgent script to open it in your code editor.
  2. In the editor, add the using Unity.MLAgents; and using Unity.MLAgents.Sensors statements and then change the base class from MonoBehaviour to Agent.
  3. Delete the Update() method, but we will use the Start() function, so leave it alone for now.

So far, these are the basic steps that you would use to add ML-Agents to any Unity project. Next, we will add the logic that will let our Agent learn to roll to the cube using reinforcement learning. More specifically, we will need to extend three methods from the Agent base class:

  • OnEpisodeBegin()
  • CollectObservations(VectorSensor sensor)
  • OnActionReceived(float[] vectorAction)

We overview each of these in more detail in the dedicated subsections below.

Initialization and Resetting the Agent

The process of training in the ML-Agents Toolkit involves running episodes where the Agent (Sphere) attempts to solve the task. Each episode lasts until the Agents solves the task (i.e. reaches the cube), fails (rolls off the platform) or times out (takes too long to solve or fail at the task). At the start of each episode, the OnEpisodeBegin() method is called to set-up the environment for a new episode. Typically the scene is initialized in a random manner to enable the agent to learn to solve the task under a variety of conditions.

In this example, each time the Agent (Sphere) reaches its target (Cube), its episode ends and the method moves the target (Cube) to a new random location. In addition, if the Agent rolls off the platform, the OnEpisodeBegin() method puts it back onto the floor.

To move the target (Cube), we need a reference to its Transform (which stores a GameObject's position, orientation and scale in the 3D world). To get this reference, add a public field of type Transform to the RollerAgent class. Public fields of a component in Unity get displayed in the Inspector window, allowing you to choose which GameObject to use as the target in the Unity Editor.

To reset the Agent's velocity (and later to apply force to move the agent) we need a reference to the Rigidbody component. A Rigidbody is Unity's primary element for physics simulation. (See Physics for full documentation of Unity physics.) Since the Rigidbody component is on the same GameObject as our Agent script, the best way to get this reference is using GameObject.GetComponent<T>(), which we can call in our script's Start() method.

So far, our RollerAgent script looks like:

using System.Collections.Generic;
using UnityEngine;
using Unity.MLAgents;
using Unity.MLAgents.Sensors;

public class RollerAgent : Agent
{
    Rigidbody rBody;
    void Start () {
        rBody = GetComponent<Rigidbody>();
    }

    public Transform Target;
    public override void OnEpisodeBegin()
    {
        if (this.transform.localPosition.y < 0)
        {
            // If the Agent fell, zero its momentum
            this.rBody.angularVelocity = Vector3.zero;
            this.rBody.velocity = Vector3.zero;
            this.transform.localPosition = new Vector3( 0, 0.5f, 0);
        }

        // Move the target to a new spot
        Target.localPosition = new Vector3(Random.value * 8 - 4,
                                           0.5f,
                                           Random.value * 8 - 4);
    }
}

Next, let's implement the Agent.CollectObservations(VectorSensor sensor) method.

Observing the Environment

The Agent sends the information we collect to the Brain, which uses it to make a decision. When you train the Agent (or use a trained model), the data is fed into a neural network as a feature vector. For an Agent to successfully learn a task, we need to provide the correct information. A good rule of thumb for deciding what information to collect is to consider what you would need to calculate an analytical solution to the problem.

In our case, the information our Agent collects includes the position of the target, the position of the agent itself, and the velocity of the agent. This helps the Agent learn to control its speed so it doesn't overshoot the target and roll off the platform. In total, the agent observation contains 8 values as implemented below:

public override void CollectObservations(VectorSensor sensor)
{
    // Target and Agent positions
    sensor.AddObservation(Target.localPosition);
    sensor.AddObservation(this.transform.localPosition);

    // Agent velocity
    sensor.AddObservation(rBody.velocity.x);
    sensor.AddObservation(rBody.velocity.z);
}

Taking Actions and Assigning Rewards

The final part of the Agent code is the Agent.OnActionReceived() method, which receives actions and assigns the reward.

Actions

To solve the task of moving towards the target, the Agent (Sphere) needs to be able to move in the x and z directions. As such, we will provide 2 actions to the agent. The first determines the force applied along the x-axis; the second determines the force applied along the z-axis. (If we allowed the Agent to move in three dimensions, then we would need a third action.

The RollerAgent applies the values from the action[] array to its Rigidbody component, rBody, using the Rigidbody.AddForce function:

Vector3 controlSignal = Vector3.zero;
controlSignal.x = action[0];
controlSignal.z = action[1];
rBody.AddForce(controlSignal * forceMultiplier);

Rewards

Reinforcement learning requires rewards. Assign rewards in the OnActionReceived() function. The learning algorithm uses the rewards assigned to the Agent during the simulation and learning process to determine whether it is giving the Agent the optimal actions. You want to reward an Agent for completing the assigned task. In this case, the Agent is given a reward of 1.0 for reaching the Target cube.

The RollerAgent calculates the distance to detect when it reaches the target. When it does, the code calls the Agent.SetReward() method to assign a reward of 1.0 and marks the agent as finished by calling the EndEpisode() method on the Agent.

float distanceToTarget = Vector3.Distance(this.transform.localPosition, Target.localPosition);
// Reached target
if (distanceToTarget < 1.42f)
{
    SetReward(1.0f);
    EndEpisode();
}

Finally, if the Agent falls off the platform, end the episode so that it can reset itself:

// Fell off platform
if (this.transform.localPosition.y < 0)
{
    EndEpisode();
}

OnActionReceived()

With the action and reward logic outlined above, the final version of the OnActionReceived() function looks like:

public float forceMultiplier = 10;
public override void OnActionReceived(float[] vectorAction)
{
    // Actions, size = 2
    Vector3 controlSignal = Vector3.zero;
    controlSignal.x = vectorAction[0];
    controlSignal.z = vectorAction[1];
    rBody.AddForce(controlSignal * forceMultiplier);

    // Rewards
    float distanceToTarget = Vector3.Distance(this.transform.localPosition, Target.localPosition);

    // Reached target
    if (distanceToTarget < 1.42f)
    {
        SetReward(1.0f);
        EndEpisode();
    }

    // Fell off platform
    if (this.transform.localPosition.y < 0)
    {
        EndEpisode();
    }
}

Note the forceMultiplier class variable is defined before the function. Since forceMultiplier is public, you can set the value from the Inspector window.

Testing the Environment

It is always a good idea to first test your environment by controlling the Agent using the keyboard. To do so, you will need to extend the Heuristic() method in the RollerAgent class. For our example, the heuristic will generate an action corresponding to the values of the "Horizontal" and "Vertical" input axis (which correspond to the keyboard arrow keys):

public override void Heuristic(float[] actionsOut)
{
    actionsOut[0] = Input.GetAxis("Horizontal");
    actionsOut[1] = Input.GetAxis("Vertical");
}

In order for the Agent to use the Heuristic, You will need to set the Behavior Type to Heuristic Only in the Behavior Parameters of the RollerAgent.

Press Play to run the scene and use the arrows keys to move the Agent around the platform. Make sure that there are no errors displayed in the Unity Editor Console window and that the Agent resets when it reaches its target or falls from the platform.

Final Editor Setup

Now, that all the GameObjects and ML-Agent components are in place, it is time to connect everything together in the Unity Editor. This involves changing some of the Agent Component's properties so that they are compatible with our Agent code.

  1. Select the RollerAgent GameObject to show its properties in the Inspector window.
  2. Add the Decision Requester script with the Add Component button from the RollerAgent Inspector.
  3. Change Decision Period to 10. For more information on decisions, see the Agent documentation
  4. Drag the Target GameObject from the Hierarchy window to the RollerAgent Target field.
  5. Add the Behavior Parameters script with the Add Component button from the RollerAgent Inspector.
  6. Modify the Behavior Parameters of the Agent :
    • Behavior Name to RollerBall
    • Vector Observation > Space Size = 8
    • Vector Action > Space Type = Continuous
    • Vector Action > Space Size = 2

Now you are ready to test the environment before training.

Training the Environment

The process is the same as described in the Getting Started Guide.

The hyperparameters for training are specified in a configuration file that you pass to the mlagents-learn program. Create a new rollerball_config.yaml file and include the following hyperparameter values:

behaviors:
  RollerBall:
    trainer_type: ppo
    hyperparameters:
      batch_size: 10
      buffer_size: 100
      learning_rate: 3.0e-4
      beta: 5.0e-4
      epsilon: 0.2
      lambd: 0.99
      num_epoch: 3
      learning_rate_schedule: linear
    network_settings:
      normalize: false
      hidden_units: 128
      num_layers: 2
    reward_signals:
      extrinsic:
        gamma: 0.99
        strength: 1.0
    max_steps: 500000
    time_horizon: 64
    summary_freq: 10000

Hyperparameters are explained in the training configuration file documentation

Since this example creates a very simple training environment with only a few inputs and outputs, using small batch and buffer sizes speeds up the training considerably. However, if you add more complexity to the environment or change the reward or observation functions, you might also find that training performs better with different hyperparameter values. In addition to setting these hyperparameter values, the Agent DecisionFrequency parameter has a large effect on training time and success. A larger value reduces the number of decisions the training algorithm has to consider and, in this simple environment, speeds up training.

To train your agent, run the following command before pressing Play in the Editor:

mlagents-learn config/rollerball_config.yaml --run-id=RollerBall

To monitor the statistics of Agent performance during training, use TensorBoard.

TensorBoard statistics display

In particular, the cumulative_reward and value_estimate statistics show how well the Agent is achieving the task. In this example, the maximum reward an Agent can earn is 1.0, so these statistics approach that value when the Agent has successfully solved the problem.

Optional: Multiple Training Areas within the Same Scene

In many of the example environments, many copies of the training area are instantiated in the scene. This generally speeds up training, allowing the environment to gather many experiences in parallel. This can be achieved simply by instantiating many Agents with the same Behavior Name. Note that we've already simplified our transition to using multiple areas by creating the TrainingArea GameObject and relying on local positions in RollerAgent.cs. Use the following steps to parallelize your RollerBall environment:

  1. Drag the TrainingArea GameObject, along with its attached GameObjects, into your Assets browser, turning it into a prefab.
  2. You can now instantiate copies of the TrainingArea prefab. Drag them into your scene, positioning them so that they do not overlap.