Unity 2018 Artificial Intelligence Cookbook: Over 90 recipes to build and customize AI entities for your games with Unity , Second Edition

Our first step is to remember the update functions' order of execution:

We need to create three classes, Steering, AgentBehaviour, and Agent:

using UnityEngine; 
public class Steering 
{ 
  public float angular; 
  public Vector3 linear; 
  public Steering () 
  { 
    angular = 0.0f; 
    linear = new Vector3(); 
  } 
} 

using UnityEngine; 
public class AgentBehaviour : MonoBehaviour 
{ 
  public GameObject target; 
  protected Agent agent; 
  public virtual void Awake () 
  { 
    agent = gameObject.GetComponent<Agent>(); 
  } 
  public virtual void Update () 
  { 
      agent.SetSteering(GetSteering()); 
  } 
  public virtual Steering GetSteering () 
  { 
    return new Steering(); 
  } 
} 

using UnityEngine; 
using System.Collections; 
public class Agent : MonoBehaviour 
{ 
    public float maxSpeed; 
    public float maxAccel; 
    public float orientation; 
    public float rotation; 
    public Vector3 velocity; 
    protected Steering steering; 
    void Start () 
    { 
        velocity = Vector3.zero; 
        steering = new Steering(); 
    } 
    public void SetSteering (Steering steering) 
    { 
        this.steering = steering; 
    } 
} 

public virtual void Update () 
{ 
    Vector3 displacement = velocity * Time.deltaTime; 
    orientation += rotation * Time.deltaTime; 
    // we need to limit the orientation values 
    // to be in the range (0 - 360) 
    if (orientation < 0.0f) 
        orientation += 360.0f; 
    else if (orientation > 360.0f) 
        orientation -= 360.0f; 
    transform.Translate(displacement, Space.World); 
    transform.rotation = new Quaternion(); 
    transform.Rotate(Vector3.up, orientation); 
} 

public virtual void LateUpdate () 
{ 
    velocity += steering.linear * Time.deltaTime; 
    rotation += steering.angular * Time.deltaTime; 
    if (velocity.magnitude > maxSpeed) 
    { 
        velocity.Normalize(); 
        velocity = velocity * maxSpeed; 
    } 
    if (steering.angular == 0.0f) 
    { 
        rotation = 0.0f; 
    } 
    if (steering.linear.sqrMagnitude == 0.0f) 
    { 
        velocity = Vector3.zero; 
    } 
    steering = new Steering(); 
} 

The idea is to be able to delegate the movement's logic inside the GetSteering() function on the behaviors that we will later build, simplifying our agent's class to a main calculation based on those.

Besides, we are guaranteed to be able to set the agent's steering value before it is used thanks to Unity script and function execution orders.

This is a component-based approach, which means that we must remember to always have an Agent script attached to GameObject for the behaviors to work as expected.

For further information on Unity's game loop and the execution order of functions and scripts, please refer to the official documentation available online at these links:

We need a couple of basic behaviors called Seek and Flee; place them right after the Agent class in the scripts' execution order.

The following is the code for the Seek behavior:

using UnityEngine; 
using System.Collections; 
public class Seek : AgentBehaviour 
{ 
    public override Steering GetSteering() 
    { 
        Steering steering = new Steering(); 
        steering.linear = target.transform.position - transform.position; 
        steering.linear.Normalize(); 
        steering.linear = steering.linear * agent.maxAccel; 
        return steering; 
    } 
} 

Also, we need to implement the Flee behavior:

using UnityEngine; 
using System.Collections; 
public class Flee : AgentBehaviour 
{ 
    public override Steering GetSteering() 
    { 
        Steering steering = new Steering(); 
        steering.linear = transform.position - target.transform.position; 
        steering.linear.Normalize(); 
        steering.linear = steering.linear * agent.maxAccel; 
        return steering; 
    } 
}

Pursue and Evade are essentially the same algorithm, but differ in terms of the base class they derive from:

using UnityEngine; 
using System.Collections; 
 
public class Pursue : Seek 
{ 
    public float maxPrediction; 
    private GameObject targetAux; 
    private Agent targetAgent; 
} 

public override void Awake() { base.Awake(); targetAgent = target.GetComponent<Agent>(); targetAux = target; target = new GameObject(); } 

void OnDestroy () 
{ 
    Destroy(targetAux); 
} 

public override Steering GetSteering() 
{ 
    Vector3 direction = targetAux.transform.position - transform.position; 
    float distance = direction.magnitude; 
    float speed = agent.velocity.magnitude; 
    float prediction; 
    if (speed <= distance / maxPrediction) 
        prediction = maxPrediction; 
    else 
        prediction = distance / speed; 
    target.transform.position = targetAux.transform.position; 
    target.transform.position += targetAgent.velocity * prediction; 
    return base.GetSteering(); 
} 

public class Evade : Flee 
{ 
    // everything stays the same 
} 

These behaviors rely on Seek and Flee and take into consideration the target's velocity in order to predict where it will go next, and they aim at that position using an internal extra object.

Our first step is to remember the execution order of event functions, but now we also consider FixedUpdate, because we're handling behaviors on top of the physics engine:

This recipe entails adding changes to our Agent class.

private Rigidbody aRigidBody; 

aRigidBody = GetComponent<Rigidbody>(); 

public Vector3 OriToVec(float orientation) 
{ 
  Vector3 vector = Vector3.zero; 
  vector.x = Mathf.Sin(orientation * Mathf.Deg2Rad) * 1.0f; 
  vector.z = Mathf.Cos(orientation * Mathf.Deg2Rad) * 1.0f; 
  return vector.normalized; 
} 

public virtual void Update () 
{ 
  if (aRigidBody == null) 
    return; 
  // ... previous code 

public virtual void FixedUpdate() 
{ 
  if (aRigidBody == null) 
    return; 
  // next step 
} 

Vector3 displacement = velocity * Time.deltaTime; 
orientation += rotation * Time.deltaTime; 
if (orientation < 0.0f) 
  orientation += 360.0f; 
else if (orientation > 360.0f) 
  orientation -= 360.0f; 
// The ForceMode will depend on what you want to achieve 
// We are using VelocityChange for illustration purposes 
aRigidBody.AddForce(displacement, ForceMode.VelocityChange); 
Vector3 orientationVector = OriToVec(orientation); 
aRigidBody.rotation = Quaternion.LookRotation(orientationVector, Vector3.up); 

We added a member variable for storing the reference to a possible rigid body component, and also implemented FixedUpdate, similar to Update, but taking into consideration that we need to apply force to the rigid body instead of translating the object ourselves, because we're working on top of Unity's physics engine.

Finally, we created a simple validation at the beginning of each function so they're called only when it applies.

For further information on the execution order of event functions, please refer to official Unity documentation:

We need to create one file for the Arrive and Leave algorithms respectively, and remember to set their custom execution order.

They use the same approach, but in terms of implementation, the name of the member variables change, as well as some computations in the first half of the GetSteering function:

using UnityEngine; 
using System.Collections; 
 
public class Arrive : AgentBehaviour 
{ 
    public float targetRadius; 
    public float slowRadius; 
    public float timeToTarget = 0.1f; 
} 

public override Steering GetSteering() 
{ 
    // code in next steps 
} 

Steering steering = new Steering(); 
Vector3 direction = target.transform.position - transform.position; 
float distance = direction.magnitude; 
float targetSpeed; 
if (distance < targetRadius) 
    return steering; 
if (distance > slowRadius) 
    targetSpeed = agent.maxSpeed; 
else 
    targetSpeed = agent.maxSpeed * distance / slowRadius; 

Vector3 desiredVelocity = direction; 
desiredVelocity.Normalize(); 
desiredVelocity *= targetSpeed; 
steering.linear = desiredVelocity - agent.velocity; 
steering.linear /= timeToTarget; 
if (steering.linear.magnitude > agent.maxAccel) 
{ 
    steering.linear.Normalize(); 
    steering.linear *= agent.maxAccel; 
} 
return steering; 

using UnityEngine; 
using System.Collections; 
 
public class Leave : AgentBehaviour 
{ 
    public float escapeRadius; 
    public float dangerRadius; 
    public float timeToTarget = 0.1f; 
} 

Steering steering = new Steering(); 
Vector3 direction = transform.position - target.transform.position; 
float distance = direction.magnitude; 
if (distance > dangerRadius) 
    return steering; 
float reduce; 
if (distance < escapeRadius) 
    reduce = 0f; 
else 
    reduce = distance / dangerRadius * agent.maxSpeed; 
float targetSpeed = agent.maxSpeed - reduce; 

After calculating the direction to go in, the next calculations are based on two radii distances in order to know when to go full throttle, slow down, and stop; that's why we have several if statements. In the Arrive behavior, when the agent is too far, we aim for full throttle, progressively slow down when inside the proper radius, and finally stop when close enough to the target. The inverse method applies to Leave:

A visual reference for the Arrive and Leave behaviors

We need to make some modifications to our AgentBehaviour class:

public float maxSpeed; 
public float maxAccel; 
public float maxRotation; 
public float maxAngularAccel;

public float MapToRange (float rotation) { 
    rotation %= 360.0f; 
    if (Mathf.Abs(rotation) > 180.0f) { 
        if (rotation < 0.0f) 
            rotation += 360.0f; 
        else 
            rotation -= 360.0f; 
    } 
    return rotation; 
} 

using UnityEngine; 
using System.Collections; 
 
public class Align : AgentBehaviour 
{ 
    public float targetRadius; 
    public float slowRadius; 
    public float timeToTarget = 0.1f; 
 
    public override Steering GetSteering() 
    { 
        Steering steering = new Steering(); 
        float targetOrientation = target.GetComponent<Agent>().orientation; 
        float rotation = targetOrientation - agent.orientation; 
        rotation = MapToRange(rotation); 
        float rotationSize = Mathf.Abs(rotation); 
        if (rotationSize < targetRadius) 
            return steering; 
        float targetRotation; 
        if (rotationSize > slowRadius) 
            targetRotation = agent.maxRotation; 
        else 
            targetRotation = agent.maxRotation * rotationSize / slowRadius; 
        targetRotation *= rotation / rotationSize; 
        steering.angular = targetRotation - agent.rotation; 
        steering.angular /= timeToTarget; 
        float angularAccel = Mathf.Abs(steering.angular); 
        if (angularAccel > agent.maxAngularAccel) 
        { 
            steering.angular /= angularAccel; 
            steering.angular *= agent.maxAngularAccel; 
        } 
        return steering; 
    } 
} 

We can now proceed to implement our facing algorithm that derives from Align:

using UnityEngine; 
using System.Collections; 
 
public class Face : Align 
{ 
    protected GameObject targetAux; 
} 

public override void Awake() 
{ 
    base.Awake(); 
    targetAux = target; 
    target = new GameObject(); 
    target.AddComponent<Agent>(); 
} 

void OnDestroy () 
{ 
    Destroy(target); 
}

public override Steering GetSteering() 
{ 
    Vector3 direction = targetAux.transform.position - transform.position; 
    if (direction.magnitude > 0.0f) 
    { 
        float targetOrientation = Mathf.Atan2(direction.x, direction.z); 
        targetOrientation *= Mathf.Rad2Deg; 
        target.GetComponent<Agent>().orientation = targetOrientation; 
    } 
    return base.GetSteering(); 
}

The algorithm computes the internal target orientation according to the vector between the agent and the real target. Then, it just delegates the work to its parent class.

We need to add another function to our AgentBehaviour class called OriToVec, which converts an orientation value to a vector:

public Vector3 GetOriAsVec (float orientation) { 
    Vector3 vector  = Vector3.zero; 
    vector.x = Mathf.Sin(orientation * Mathf.Deg2Rad) * 1.0f; 
    vector.z = Mathf.Cos(orientation * Mathf.Deg2Rad) * 1.0f; 
    return vector.normalized; 
} 

We can regard it as a big three-step process in which we first manipulate the internal target position in a parameterized random way, face that position, and move accordingly:

using UnityEngine; 
using System.Collections; 
 
public class Wander : Face 
{ 
    public float offset; 
    public float radius; 
    public float rate; 
} 

public override void Awake() 
{ 
    target = new GameObject(); 
    target.transform.position = transform.position; 
    base.Awake(); 
} 

public override Steering GetSteering() 
{ 
    Steering steering = new Steering(); 
    float wanderOrientation = Random.Range(-1.0f, 1.0f) * rate; 
    float targetOrientation = wanderOrientation + agent.orientation; 
    Vector3 orientationVec = OriToVec(agent.orientation); 
    Vector3 targetPosition = (offset * orientationVec) + transform.position; 
    targetPosition = targetPosition + (OriToVec(targetOrientation) * radius); 
    targetAux.transform.position = targetPosition; 
    steering = base.GetSteering(); 
    steering.linear = targetAux.transform.position - transform.position; 
    steering.linear.Normalize(); 
    steering.linear *= agent.maxAccel; 
    return steering; 
} 

The behavior takes two radii in order into consideration to get a random position to go next, looks toward that random point, and converts the computed orientation into a direction vector in order to advance:

A visual description of the parameters for creating the Wander behavior

We need to define a custom data type called PathSegment:

using UnityEngine; 
using System.Collections; 
 
public class PathSegment 
{ 
    public Vector3 a; 
    public Vector3 b; 
 
    public PathSegment () : this (Vector3.zero, Vector3.zero){} 
    public PathSegment (Vector3 a, Vector3 b) 
    { 
        this.a = a; 
        this.b = b; 
    } 
} 

This is a long recipe that could be regarded as a big two-step process. First, we build the Path class, which abstracts points in the path from their specific spatial representations, and then we build the PathFollower behavior that makes use of that abstraction in order to get actual spatial points to follow:

using UnityEngine; 
using System.Collections; 
using System.Collections.Generic; 
 
public class Path : MonoBehaviour 
{ 
    public List<GameObject> nodes; 
    List<PathSegment> segments; 
} 

void Start() 
{ 
    segments = GetSegments(); 
} 

public List<PathSegment> GetSegments () 
{ 
    List<PathSegment> segments = new List<PathSegment>(); 
    int i; 
    for (i = 0; i < nodes.Count - 1; i++) 
    { 
        Vector3 src = nodes[i].transform.position; 
        Vector3 dst = nodes[i+1].transform.position; 
        PathSegment segment = new PathSegment(src, dst); 
        segments.Add(segment); 
    } 
    return segments; 
} 

public float GetParam(Vector3 position, float lastParam) 
{ 
    // body 
} 

float param = 0f; 
PathSegment currentSegment = null; 
float tempParam = 0f; 
foreach (PathSegment ps in segments) 
{ 
    tempParam += Vector3.Distance(ps.a, ps.b); 
    if (lastParam <= tempParam) 
    { 
        currentSegment = ps; 
        break; 
    } 
} 
if (currentSegment == null) 
    return 0f; 

Vector3 currPos = position - currentSegment.a; 
Vector3 segmentDirection = currentSegment.b - currentSegment.a; 
segmentDirection.Normalize(); 

Vector3 pointInSegment = Vector3.Project(currPos, segmentDirection); 

param = tempParam - Vector3.Distance(currentSegment.a, currentSegment.b);

param += pointInSegment.magnitude; 
return param;

public Vector3 GetPosition(float param)  
{ 
    // body 
} 

Vector3 position = Vector3.zero; 
PathSegment currentSegment = null; 
float tempParam = 0f; 
foreach (PathSegment ps in segments) 
{ 
    tempParam += Vector3.Distance(ps.a, ps.b); 
    if (param <= tempParam) 
    { 
        currentSegment = ps; 
        break; 
    } 
} 
if (currentSegment == null) 
    return Vector3.zero; 

Vector3 segmentDirection = currentSegment.b - currentSegment.a; 
segmentDirection.Normalize(); 
tempParam -= Vector3.Distance(currentSegment.a, currentSegment.b); 
tempParam = param - tempParam; 
position = currentSegment.a + segmentDirection * tempParam; 
return position; 

using UnityEngine; 
using System.Collections; 
 
public class PathFollower : Seek 
{ 
    public Path path; 
    public float pathOffset = 0.0f; 
    float currentParam; 
}

public override void Awake() 
{ 
    base.Awake(); 
    target = new GameObject(); 
    currentParam = 0f; 
} 

public override Steering GetSteering() 
{ 
    currentParam = path.GetParam(transform.position, currentParam); 
    float targetParam = currentParam + pathOffset; 
    target.transform.position = path.GetPosition(targetParam); 
    return base.GetSteering(); 
} 

We use the Path class in order to have a movement guideline. It is the cornerstone, because it relies on GetParam to map an offset point to follow in its internal guideline, and it also uses GetPosition to convert that referential point to a position in the three-dimensional space along the segments.

The path-following algorithm just makes use of the path's functions in order to get a new position, update the target, and apply the Seek behavior.

It's important to take into account the order in which the nodes are linked in the inspector for the path to work as expected. A practical way to achieve this is to manually name the nodes with a reference number:

An example of a path set up in the Inspector window

Also, we could define the OnDrawGizmos function in order to have a better visual reference of the path:

void OnDrawGizmos () 
{ 
    Vector3 direction; 
    Color tmp = Gizmos.color; 
    Gizmos.color = Color.magenta;//example color 
    int i; 
    for (i = 0; i < nodes.Count - 1; i++) 
    { 
        Vector3 src = nodes[i].transform.position; 
        Vector3 dst = nodes[i+1].transform.position; 
        direction = dst - src; 
        Gizmos.DrawRay(src, direction); 
    } 
    Gizmos.color = tmp; 
} 

We need to create a tag called Agent and assign it to those game objects that we would like to avoid, and we also need to have the Agent script component attached to them:

Example of how the Inspector window of a dummy agent should appear

Take a look on the following:

This recipe will entail creating a new agent behavior:

using UnityEngine; 
using System.Collections; 
using System.Collections.Generic; 
 
public class AvoidAgent : AgentBehaviour 
{ 
    public float collisionRadius = 0.4f; 
    GameObject[] targets; 
} 

void Start () 
{ 
    targets = GameObject.FindGameObjectsWithTag("Agent"); 
} 

public override Steering GetSteering() 
{ 
    // body 
} 

Steering steering = new Steering(); 
float shortestTime = Mathf.Infinity; 
GameObject firstTarget = null; 
float firstMinSeparation = 0.0f; 
float firstDistance = 0.0f; 
Vector3 firstRelativePos = Vector3.zero; 
Vector3 firstRelativeVel = Vector3.zero; 

foreach (GameObject t in targets) 
{ 
    Vector3 relativePos; 
    Agent targetAgent = t.GetComponent<Agent>(); 
    relativePos = t.transform.position - transform.position; 
    Vector3 relativeVel = targetAgent.velocity - agent.velocity; 
    float relativeSpeed = relativeVel.magnitude; 
    float timeToCollision = Vector3.Dot(relativePos, relativeVel); 
    timeToCollision /= relativeSpeed * relativeSpeed * -1; 
    float distance = relativePos.magnitude; 
    float minSeparation = distance - relativeSpeed * timeToCollision; 
    if (minSeparation > 2 * collisionRadius) 
        continue; 
    if (timeToCollision > 0.0f && timeToCollision < shortestTime) 
    { 
        shortestTime = timeToCollision; 
        firstTarget = t; 
        firstMinSeparation = minSeparation; 
        firstRelativePos = relativePos; 
        firstRelativeVel = relativeVel; 
    } 
} 

if (firstTarget == null) 
    return steering; 
if (firstMinSeparation <= 0.0f || firstDistance < 2 * collisionRadius) 
    firstRelativePos = firstTarget.transform.position; 
else 
    firstRelativePos += firstRelativeVel * shortestTime; 
firstRelativePos.Normalize(); 
steering.linear = -firstRelativePos * agent.maxAccel; 
return steering; 

Given a list of agents, we take into consideration which one is closest, and, if it is close enough, we make it so that the agent tries to escape from the expected route of that first one according to its current velocity, so that they don't collide.

This behavior works well when combined with other behaviors using blending techniques (some are included in this chapter); otherwise, it's a starting point for your own collision avoidance algorithms.

This technique uses the RaycastHit structure and the Raycast function from the physics engine, so it's recommended that you look at the documents for a refresher in case you're a little rusty on the subject.

Thanks to our previous hard work, this recipe is a short one:

using UnityEngine; 
using System.Collections; 
 
public class AvoidWall : Seek 
{ 
    // body 
} 

public float avoidDistance; 
public float lookAhead; 

public override void Awake() 
{ 
    base.Awake(); 
    target = new GameObject(); 
} 

public override Steering GetSteering() 
{ 
    // body 
}

Steering steering = new Steering(); 
Vector3 position = transform.position; 
Vector3 rayVector = agent.velocity.normalized * lookAhead; 
Vector3 direction = rayVector; 
RaycastHit hit; 

if (Physics.Raycast(position, direction, out hit, lookAhead)) 
{ 
    position = hit.point + hit.normal * avoidDistance; 
    target.transform.position = position; 
    steering = base.GetSteering(); 
} 
return steering; 

We cast a ray in front of the agent, and when the ray collides with a wall, the target object is placed in a new position, with consideration given to its distance from the wall and the safety distance declared, delegating the steering calculations to the Seek behavior; this creates the illusion of the agent avoiding the wall.

We could extend this behavior by adding more rays, such as whiskers, in order to achieve better accuracy. Also, it is usually paired with other movement behaviors, such as Pursue, using blending:

The original ray cast and possible extensions for more precise wall avoidance

For further information on the RaycastHit structure and the Raycast function, please refer to the official documentation available online at these links:

We must add a new member variable to our AgentBehaviour class called weight and preferably assign a default value. In this case, this is 1.0f. We should refactor the Update function to incorporate weight as a parameter for the Agent class's SetSteering function. All in all, the new AgentBehaviour class should look something like this:

public class AgentBehaviour : MonoBehaviour 
{ 
    public float weight = 1.0f; 
 
    // ... the rest of the class 
 
    public virtual void Update () 
    { 
        agent.SetSteering(GetSteering(), weight); 
   } 
}

We just need to change the SetSteering function's signature and definition:

public void SetSteering (Steering steering, float weight) 
{ 
    this.steering.linear += (weight * steering.linear); 
    this.steering.angular += (weight * steering.angular); 
} 

The weights are used to amplify the steering behavior result, and they're added to the main steering structure.

The weights don't necessarily need to add up to 1.0f. The weight parameter is a reference for defining the relevance that the steering behavior will have among other parameters.

In this project, there is an example of avoiding walls, which is worked out using weighted blending.

This approach is very similar to the one used in the previous recipe. We must add a new member variable to our AgentBehaviour class. We should also refactor the Update function to incorporate priority as a parameter to the Agent class's SetSteering function. The new AgentBehaviour class should look something like this:

public class AgentBehaviour : MonoBehaviour 
{ 
    public int priority = 1; 
    // ... everything else stays the same 
    public virtual void Update () 
    { 
        agent.SetSteering(GetSteering(), priority); 
    } 
} 

Now, we need to make some changes to the Agent class:

using System.Collections.Generic; 

public float priorityThreshold = 0.2f; 

private Dictionary<int, List<Steering>> groups; 

groups = new Dictionary<int, List<Steering>>(); 

public virtual void LateUpdate () 
{ 
    //  funnelled steering through priorities 
    steering = GetPrioritySteering(); 
    groups.Clear(); 
    // ... the rest of the computations stay the same 
    steering = new Steering(); 
} 

public void SetSteering (Steering steering, int priority) 
{ 
    if (!groups.ContainsKey(priority)) 
    { 
        groups.Add(priority, new List<Steering>()); 
    } 
    groups[priority].Add(steering); 
} 

private Steering GetPrioritySteering () 
{ 
    Steering steering = new Steering(); 
    float sqrThreshold = priorityThreshold * priorityThreshold; 
    foreach (List<Steering> group in groups.Values) 
    { 
        steering = new Steering(); 
        foreach (Steering singleSteering in group) 
        { 
            steering.linear += singleSteering.linear; 
            steering.angular += singleSteering.angular; 
        } 
        if (steering.linear.sqrMagnitude > sqrThreshold || 
                Mathf.Abs(steering.angular) > priorityThreshold) 
        { 
            return steering; 
        } 
    } 
    return steering; 
}

By creating priority groups, we blend behaviors that are common to one another, and the first group, in which the steering value exceeds the threshold, is selected. Otherwise, steering from the lowest-priority group is chosen.

We could extend this approach by mixing it with weighted blending, so we would have a more robust architecture, achieving extra precision in the way the behaviors impact on the agent at every priority level:

foreach (Steering singleSteering in group) 
{ 
    steering.linear += singleSteering.linear * weight; 
    steering.angular += singleSteering.angular * weight; 
} 

There is an example of avoiding walls using priority-based blending in this project.

This recipe differs a little bit as it doesn't rely on the base AgentBehaviour class.

using UnityEngine; 
using System.Collections; 
 
public class Projectile : MonoBehaviour 
{ 
    private bool set = false; 
    private Vector3 firePos; 
    private Vector3 direction; 
    private float speed; 
    private float timeElapsed; 
} 

void Update () 
{ 
    if (!set) 
        return; 
    timeElapsed += Time.deltaTime; 
    transform.position = firePos + direction * speed * timeElapsed; 
    transform.position += Physics.gravity * (timeElapsed * timeElapsed) / 2.0f; 
    // extra validation for cleaning the scene 
    if (transform.position.y < -1.0f) 
        Destroy(this.gameObject);// or set = false; and hide it 
} 

public void Set (Vector3 firePos, Vector3 direction, float speed) 
{ 
    this.firePos = firePos; 
    this.direction = direction.normalized; 
    this.speed = speed; 
    transform.position = firePos; 
    set = true; 
} 

This behavior uses high school physics in order to generate the parabolic movement.

We could also take another approach: implementing public properties in the script or declaring member variables as public, and instead of calling the Set function, having the script disabled by default in the prefab and enabling it after all the properties have been set. That way, we could easily apply the object pool pattern.

For further information on the object pool pattern, please refer to the following Wikipedia article and the official Unity technologies video tutorial available online:

Before we get into predicting the landing position, it's important to know the time left before it hits the ground (or reaches a certain position). Thus, instead of creating new behaviors, we need to update the Projectile class.

public float GetLandingTime (float height = 0.0f) 
{ 
    Vector3 position = transform.position; 
    float time = 0.0f; 
    float valueInt = (direction.y * direction.y) * (speed * speed); 
    valueInt = valueInt - (Physics.gravity.y * 2 * (position.y - height)); 
    valueInt = Mathf.Sqrt(valueInt); 
    float valueAdd = (-direction.y) * speed; 
    float valueSub = (-direction.y) * speed; 
    valueAdd = (valueAdd + valueInt) / Physics.gravity.y; 
    valueSub = (valueSub - valueInt) / Physics.gravity.y; 
    if (float.IsNaN(valueAdd) && !float.IsNaN(valueSub)) 
        return valueSub; 
    else if (!float.IsNaN(valueAdd) && float.IsNaN(valueSub)) 
        return valueAdd; 
    else if (float.IsNaN(valueAdd) && float.IsNaN(valueSub)) 
        return -1.0f; 
    time = Mathf.Max(valueAdd, valueSub); 
    return time; 
} 

public Vector3 GetLandingPos (float height = 0.0f) 
{ 
    Vector3 landingPos = Vector3.zero; 
    float time = GetLandingTime(); 
    if (time < 0.0f) 
        return landingPos; 
    landingPos.y = height; 
    landingPos.x = firePos.x + direction.x * speed * time; 
    landingPos.z = firePos.z + direction.z * speed * time; 
    return landingPos; 
}

First, we are solving the equation from the previous recipe for a fixed height, and, given the projectile's current position and speed, we are able to get the time at which the projectile will reach the given height.

Remember to take into account the NaN validation. It's placed that way because there may be one, two, or no solutions to the equation. Furthermore, when the landing time is less than zero, it means the projectile won't be able to reach the target height.

Just as in the previous recipe, we only need to expand the Projectile class.

Thanks to our previous hard work, this recipe is a real piece of cake:

public static Vector3 GetFireDirection (Vector3 startPos, Vector3 endPos, float speed) 
{ 
    // body 
}

Vector3 direction = Vector3.zero; 
Vector3 delta = endPos - startPos; 
float a = Vector3.Dot(Physics.gravity, Physics.gravity); 
float b = -4 * (Vector3.Dot(Physics.gravity, delta) + speed * speed); 
float c = 4 * Vector3.Dot(delta, delta); 
if (4 * a * c > b * b) 
    return direction; 
float time0 = Mathf.Sqrt((-b + Mathf.Sqrt(b * b - 4 * a * c)) / (2*a)); 
float time1 = Mathf.Sqrt((-b - Mathf.Sqrt(b * b - 4 * a * c)) / (2*a)); 

float time; 
if (time0 < 0.0f) 
{ 
    if (time1 < 0) 
        return direction; 
    time = time1; 
} 
else 
{ 
    if (time1 < 0) 
        time = time0; 
    else 
        time = Mathf.Min(time0, time1); 
} 
direction = 2 * delta - Physics.gravity * (time * time); 
direction = direction / (2 * speed * time); 
return direction; 

Given a fixed speed, we solve the corresponding quadratic equation in order to obtain the desired direction (when at least a single one-time value is available), which doesn't need to be normalized because we already normalized the vector while setting up the projectile.

Take account of the fact that we are returning a blank direction when the time is negative; it means that the speed is not sufficient. One way to overcome this is to define a function that tests different speeds and then shoot the projectile.

Another relevant improvement is to add an extra parameter of the type bool for those cases when we have two valid times (which means two possible arcs), and we need to shoot over an obstacle such as a wall:

if (isWall) 
    time = Mathf.Max(time0, time1); 
else 
    time = Mathf.Min(time0, time1); 

We need to create a basic matching-velocity algorithm and the notion of jump pads and landing pads in order to emulate velocity math so that we can reach them.

The following is the code for the VelocityMatch behavior:

using UnityEngine; 
using System.Collections; 
 
public class VelocityMatch : AgentBehaviour { 
     
    public float timeToTarget = 0.1f; 
 
    public override Steering GetSteering() 
    { 
        Steering steering = new Steering(); 
        steering.linear = target.GetComponent<Agent>().velocity - agent.velocity; 
        steering.linear /= timeToTarget; 
        if (steering.linear.magnitude > agent.maxAccel) 
            steering.linear = steering.linear.normalized * agent.maxAccel; 
 
        steering.angular = 0.0f; 
        return steering; 
    } 
} 

Also, it's important to create a data type called JumpPoint:

using UnityEngine; 
using System.Collections; 
 
public class JumpPoint  
{ 
    public Vector3 jumpLocation; 
    public Vector3 landingLocation; 
 
    //The change in position from jump to landing 
    public Vector3 deltaPosition; 
 
    public JumpPoint () : this (Vector3.zero, Vector3.zero) 
    { 
    } 
 
    public JumpPoint(Vector3 a, Vector3 b) 
    { 
        this.jumpLocation = a; 
        this.landingLocation = b; 
        this.deltaPosition = this.landingLocation - this.jumpLocation; 
    } 
} 

using UnityEngine; 
using System.Collections; 
 
public class Jump : VelocityMatch 
{ 
    public JumpPoint jumpPoint; 
    public float maxYVelocity; 
    public Vector3 gravity = new Vector3(0, -9.8f, 0); 
    bool canAchieve = false; 
}

public void SetJumpPoint(Transform jumpPad, Transform landingPad) 
{ 
    jumpPoint = new JumpPoint(jumpPad.position, landingPad.position); 
} 

protected void CalculateTarget() 
{ 
    target = new GameObject(); 
    target.AddComponent<Agent>(); 
    target.transform.position = jumpPoint.jumpLocation; 
    //Calculate the first jump time 
    float sqrtTerm = Mathf.Sqrt(2f * gravity.y * jumpPoint.deltaPosition.y + maxYVelocity * agent.maxSpeed); 
    float time = (maxYVelocity - sqrtTerm) / gravity.y; 
    //Check if we can use it, otherwise try the other time 
    if (!CheckJumpTime(time)) 
    { 
        time = (maxYVelocity + sqrtTerm) / gravity.y; 
    } 
} 

private bool CheckJumpTime(float time) 
{ 
    //Calculate the planar speed 
    float vx = jumpPoint.deltaPosition.x / time; 
    float vz = jumpPoint.deltaPosition.z / time; 
    float speedSq = vx * vx + vz * vz; 
    //Check it to see if we have a valid solution 
    if (speedSq < agent.maxSpeed * agent.maxSpeed) 
    { 
        target.GetComponent<Agent>().velocity = new Vector3(vx, 0f, vz); 
        canAchieve = true; 
        return true; 
    } 
    return false; 
}

public override Steering GetSteering() 
{ 
    Steering steering = new Steering(); 
    if (target == null) 
    { 
        CalculateTarget(); 
    } 
    if (!canAchieve) 
    { 
        return steering; 
    } 
    //Check if we've hit the jump point 
    if (Mathf.Approximately((transform.position - target.transform.position).magnitude, 0f) && 
        Mathf.Approximately((agent.velocity - target.GetComponent<Agent>().velocity).magnitude, 0f)) 
    { 
        // call a jump method based on the Projectile behaviour 
        return steering; 
    } 
    return base.GetSteering(); 
} 

The algorithm takes into account the agent's velocity and calculates whether it can reach the landing pad or not. If it judges that the agent can, it tries to match the vertical velocity while seeking the landing pad's position.