ml-agents/Project/Assets/ML-Agents/Examples/Crawler/Scripts/CrawlerAgent.cs

using System;
using UnityEngine;
using Unity.MLAgents;
using Unity.MLAgentsExamples;
using Unity.MLAgents.Sensors;
using Random = UnityEngine.Random;

[RequireComponent(typeof(JointDriveController))] // Required to set joint forces
public class CrawlerAgent : Agent
{
    public float maximumWalkingSpeed = 999; //The max walk velocity magnitude an agent will be rewarded for
    Vector3 m_WalkDir; //Direction to the target
    Quaternion m_WalkDirLookRot; //Will hold the rotation to our target

    [Header("Target To Walk Towards")] [Space(10)]
    public TargetController target; //Target the agent will walk towards.

    [Header("Body Parts")] [Space(10)] public Transform body;
    public Transform leg0Upper;
    public Transform leg0Lower;
    public Transform leg1Upper;
    public Transform leg1Lower;
    public Transform leg2Upper;
    public Transform leg2Lower;
    public Transform leg3Upper;
    public Transform leg3Lower;


    [Header("Orientation")] [Space(10)]
    //This will be used as a stabilized model space reference point for observations
    //Because ragdolls can move erratically during training, using a stabilized reference transform improves learning
    public OrientationCubeController orientationCube;

    JointDriveController m_JdController;

    [Header("Reward Functions To Use")] [Space(10)]
    public bool rewardMovingTowardsTarget; // Agent should move towards target

    public bool rewardFacingTarget; // Agent should face the target
    public bool rewardUseTimePenalty; // Hurry up

    [Header("Foot Grounded Visualization")] [Space(10)]
    public bool useFootGroundedVisualization;

    public MeshRenderer foot0;
    public MeshRenderer foot1;
    public MeshRenderer foot2;
    public MeshRenderer foot3;
    public Material groundedMaterial;
    public Material unGroundedMaterial;

    public override void Initialize()
    {
        orientationCube.UpdateOrientation(body, target.transform);

        m_JdController = GetComponent<JointDriveController>();

        //Setup each body part
        m_JdController.SetupBodyPart(body);
        m_JdController.SetupBodyPart(leg0Upper);
        m_JdController.SetupBodyPart(leg0Lower);
        m_JdController.SetupBodyPart(leg1Upper);
        m_JdController.SetupBodyPart(leg1Lower);
        m_JdController.SetupBodyPart(leg2Upper);
        m_JdController.SetupBodyPart(leg2Lower);
        m_JdController.SetupBodyPart(leg3Upper);
        m_JdController.SetupBodyPart(leg3Lower);
    }

    /// <summary>
    /// Loop over body parts and reset them to initial conditions.
    /// </summary>
    public override void OnEpisodeBegin()
    {
        foreach (var bodyPart in m_JdController.bodyPartsDict.Values)
        {
            bodyPart.Reset(bodyPart);
        }

        //Random start rotation to help generalize
        transform.rotation = Quaternion.Euler(0, Random.Range(0.0f, 360.0f), 0);

        orientationCube.UpdateOrientation(body, target.transform);
    }

    /// <summary>
    /// Add relevant information on each body part to observations.
    /// </summary>
    public void CollectObservationBodyPart(BodyPart bp, VectorSensor sensor)
    {
        //GROUND CHECK
        sensor.AddObservation(bp.groundContact.touchingGround); // Is this bp touching the ground

        //Get velocities in the context of our orientation cube's space
        //Note: You can get these velocities in world space as well but it may not train as well.
        sensor.AddObservation(orientationCube.transform.InverseTransformDirection(bp.rb.velocity));
        sensor.AddObservation(orientationCube.transform.InverseTransformDirection(bp.rb.angularVelocity));

        //Get position relative to hips in the context of our orientation cube's space
        sensor.AddObservation(orientationCube.transform.InverseTransformDirection(bp.rb.position - body.position));

        if (bp.rb.transform != body)
        {
            sensor.AddObservation(bp.rb.transform.localRotation);
            sensor.AddObservation(bp.currentStrength / m_JdController.maxJointForceLimit);
        }
    }

    /// <summary>
    /// Loop over body parts to add them to observation.
    /// </summary>
    public override void CollectObservations(VectorSensor sensor)
    {
        //Add body rotation delta relative to orientation cube
        sensor.AddObservation(Quaternion.FromToRotation(body.forward, orientationCube.transform.forward));

        //Add pos of target relative to orientation cube
        sensor.AddObservation(orientationCube.transform.InverseTransformPoint(target.transform.position));

        RaycastHit hit;
        float maxRaycastDist = 10;
        if (Physics.Raycast(body.position, Vector3.down, out hit, maxRaycastDist))
        {
            sensor.AddObservation(hit.distance / maxRaycastDist);
        }
        else
            sensor.AddObservation(1);

        foreach (var bodyPart in m_JdController.bodyPartsList)
        {
            CollectObservationBodyPart(bodyPart, sensor);
        }
    }

    /// <summary>
    /// Agent touched the target
    /// </summary>
    public void TouchedTarget()
    {
        AddReward(1f);
    }

    public override void OnActionReceived(float[] vectorAction)
    {
        // The dictionary with all the body parts in it are in the jdController
        var bpDict = m_JdController.bodyPartsDict;

        var i = -1;
        // Pick a new target joint rotation
        bpDict[leg0Upper].SetJointTargetRotation(vectorAction[++i], vectorAction[++i], 0);
        bpDict[leg1Upper].SetJointTargetRotation(vectorAction[++i], vectorAction[++i], 0);
        bpDict[leg2Upper].SetJointTargetRotation(vectorAction[++i], vectorAction[++i], 0);
        bpDict[leg3Upper].SetJointTargetRotation(vectorAction[++i], vectorAction[++i], 0);
        bpDict[leg0Lower].SetJointTargetRotation(vectorAction[++i], 0, 0);
        bpDict[leg1Lower].SetJointTargetRotation(vectorAction[++i], 0, 0);
        bpDict[leg2Lower].SetJointTargetRotation(vectorAction[++i], 0, 0);
        bpDict[leg3Lower].SetJointTargetRotation(vectorAction[++i], 0, 0);

        // Update joint strength
        bpDict[leg0Upper].SetJointStrength(vectorAction[++i]);
        bpDict[leg1Upper].SetJointStrength(vectorAction[++i]);
        bpDict[leg2Upper].SetJointStrength(vectorAction[++i]);
        bpDict[leg3Upper].SetJointStrength(vectorAction[++i]);
        bpDict[leg0Lower].SetJointStrength(vectorAction[++i]);
        bpDict[leg1Lower].SetJointStrength(vectorAction[++i]);
        bpDict[leg2Lower].SetJointStrength(vectorAction[++i]);
        bpDict[leg3Lower].SetJointStrength(vectorAction[++i]);
    }

    void FixedUpdate()
    {
        orientationCube.UpdateOrientation(body, target.transform);

        // If enabled the feet will light up green when the foot is grounded.
        // This is just a visualization and isn't necessary for function
        if (useFootGroundedVisualization)
        {
            foot0.material = m_JdController.bodyPartsDict[leg0Lower].groundContact.touchingGround
                ? groundedMaterial
                : unGroundedMaterial;
            foot1.material = m_JdController.bodyPartsDict[leg1Lower].groundContact.touchingGround
                ? groundedMaterial
                : unGroundedMaterial;
            foot2.material = m_JdController.bodyPartsDict[leg2Lower].groundContact.touchingGround
                ? groundedMaterial
                : unGroundedMaterial;
            foot3.material = m_JdController.bodyPartsDict[leg3Lower].groundContact.touchingGround
                ? groundedMaterial
                : unGroundedMaterial;
        }

        // Set reward for this step according to mixture of the following elements.
        if (rewardMovingTowardsTarget)
        {
            RewardFunctionMovingTowards();
        }

        if (rewardFacingTarget)
        {
            RewardFunctionFacingTarget();
        }

        if (rewardUseTimePenalty)
        {
            RewardFunctionTimePenalty();
        }
    }

    /// <summary>
    /// Reward moving towards target & Penalize moving away from target.
    /// </summary>
    void RewardFunctionMovingTowards()
    {
        var movingTowardsDot = Vector3.Dot(orientationCube.transform.forward,
            Vector3.ClampMagnitude(m_JdController.bodyPartsDict[body].rb.velocity, maximumWalkingSpeed));
        if (float.IsNaN(movingTowardsDot))
        {
            throw new ArgumentException(
                "NaN in movingTowardsDot.\n" +
                $" orientationCube.transform.forward: {orientationCube.transform.forward}\n"+
                $" body.velocity: {m_JdController.bodyPartsDict[body].rb.velocity}\n"+
                $" maximumWalkingSpeed: {maximumWalkingSpeed}"
            );
        }
        AddReward(0.03f * movingTowardsDot);
    }

    /// <summary>
    /// Reward facing target & Penalize facing away from target
    /// </summary>
    void RewardFunctionFacingTarget()
    {
        var facingReward = Vector3.Dot(orientationCube.transform.forward, body.forward);
        if (float.IsNaN(facingReward))
        {
            throw new ArgumentException(
                "NaN in movingTowardsDot.\n" +
                $" orientationCube.transform.forward: {orientationCube.transform.forward}\n"+
                $" body.forward: {body.forward}"
            );
        }
        AddReward(0.01f * facingReward);
    }

    /// <summary>
    /// Existential penalty for time-contrained tasks.
    /// </summary>
    void RewardFunctionTimePenalty()
    {
        AddReward(-0.001f);
    }
}