Top Down Stealth Toolkit Tutorial: How to control the AI Perception model for AI agents

The following information is based on the v2.0 edition of Top Down Stealth Toolkit & hence may not remain entirely relevant in later versions. For more information about the toolkit, check out the official support thread in the Unreal Engine forums: https://forums.unrealengine.com/showthread.php?97156-Top-Down-Stealth-Toolkit


The Top Down Stealth Toolkit uses a custom Perception system to evaluate potential threats for the AI agents. It's user defined properties can all be edited through the component details panel from the owning blueprint, thus facilitating creation of different threat perception models for different types of AI agents. This tutorial will go over the basic parameters that control the working of this system.



Visual Perception Data:

1. IsVisualPerceptionEnabled?: Determines if the agent can use Visual Perception to perceive stimuli.
2. Vision Range: Determines the direct line of sight range.
3. HalfVisionAngle: Determines the half angle for the agent's cone of vision.

Aural Perception Data:

1. IsAuralPerceptionEnabled?: Determines if the agent can use Aural Perception to perceive stimuli.
2. HearingRange: Controls the maximum distance at which a sound of default loudness 1.0 (controlled through the 'PerceptionRangeModifier' parameter of interest stimuli: Breakdown of Stimulus Parameters) can be perceived by the agent.

Intuitive Perception Data:

1. IsIntuitivePerceptionEnabled?: Determines if the agent can use Intuitive Perception to perceive stimuli.

Motion Perception Data:

1. IsMotionPerceptionEnabled?: Determines if the agent can use Motion Perception to perceive a stimuli.
2. DetectionRange: Determines the maximum range at which stimuli can be sensed.
3. DetectionCounter: A counter that decides if the agent should respond to the stimulus [Used because this is an indirect form of perception]
4. CounterIncrementPerCycle: Controls the rate at which the Detection Counter increases if the stimulus is in range.
5. CounterDecrementPerCycle: Controls the rate at which the Detection Counter decreases once the stimulus goes out of range.
6. CriticalDetectionValue: Determines the threshold point for Detection Counter, at which the agent responds to the stimulus.
7. DisplayAlertnessLevel?: Determines if changes in the Detection Counter value need to be displayed in the game space.

Can Perceive Interests?: Determines if the Agent can perceive Interest Stimuli. If set to false, only Target stimuli will be perceivable.


Note: The Perception system has no awareness about it's owning actor & uses a custom interface to receive & send messages. So if you're using a custom AI agent, make sure that the owner class implements the BPI_SensorySystem interface & it's functions. For reference, check out the implementation of the interface functions in BP_PatrolGuard_Parent.

Top Down Stealth Toolkit Tutorial: How to create a new custom Interest Stimulus

The following information is based on the v2.0 edition of Top Down Stealth Toolkit & hence may not remain entirely relevant in later versions. For more information about the toolkit, check out the official support thread in the Unreal Engine forums: https://forums.unrealengine.com/showthread.php?97156-Top-Down-Stealth-Toolkit]

The v2.0 update for Top Down Stealth Toolkit introduced a dedicated stimulus driven AI perception system. This tutorial will go over the process of using the aforementioned system to create a new Interest stimulus from scratch.


The base attributes for Interest stimuli are controlled through the 'InterestStimulusDataArray' in BP_AISensoryManager class. Each element of the struct array contains details about a specific type of Interest & hence provide an avenue for easily customizing it's properties. Before going into the actual process of creating a new Interest, I'll provide a brief description of the various parameters that control that control these stimuli.

1. InterestType: Determines the type of Interest associated with the array element.

2. PerceptionType: Determines the AI perception model that can detect this Interest.

3. ThreatClass: Determines the priority level of an Interest when it comes to override requests. For example, an Interest of lower threat class will not override one of higher threat class, thus making sure that the AI retains focus on the stimuli based on their relative importance. By default, all Interests have a threat class level of 1.

4. PerceivedThreatValue: Maintains a normalized [0.0 to 1.0] value of the threat rating of an Interest as perceived by AI agents. Higher values of this parameter will make AI bots go directly into a high alert level state. Within a single cycle of sensory evaluation, Interests with higher value of this parameter within a threat class, are evaluated first to ensure that the AI responds to the most immediate threat.

5. PerceptionRangeModifier: Controls the range at which an Interest can be perceived by adding a distance modifier to the agent's perception checks. For example, Alarm & Gunshot noises have a value of 2.0, thus ensuring that an AI agent possessing a hearing range of 500 m can perceive these stimuli up to double the said distance.

6. PerformRangeTestToInterestCreator: Determines if AI perception tests should conduct distance checks against the actual Interest location, or the entity that created it. For example, perception checks for alarm noises are conducted against the actual source of the noise itself, even though the Interest location would be set as the location of the entity that triggered the alarm.

7. TrackAlertedAgents?: Determines if the Interest object should keep track of agents that have been alerted to it's presence.

8. UseInterestCreatorLocation?: Controls whether the AI agents that detected this stimulus move towards the Interest location or the actor that created it.

9. AutoDeactivate?: Controls the active lifespan of the Interest. For example, noise stimuli have this parameter turned on, thus making them relevant only for a single evaluation cycle.

10. IsInteractive?: Determines if the AI agents can interact with the Interest. If turned off, the AI bots will only investigate the source location.

11. InteractionModel: Controls the manner in which an agent interacts with an interactive Interest. For example, patrol guards will use the 'Reactivate' interaction model to wake up any incapacitated allies they may come across.

12. SortedPriorityOrder: This parameter is set automatically at the start of a game based on the threat class & perceived threat value of the Interest. Ensures that AI Sensory Manager evaluates the Interests with higher priority order before moving on to the rest.


Now in order to create a new Interest, first add a new entry to the enum 'EInterestStimulusType'. All that's left is to add an element to the InterestTypeDataArray for this new type, & assign the parameters mentioned above as per the requirements.

Top Down Stealth Toolkit Basics: AI Perception

The v2.0 update for Top Down Stealth Toolkit introduced a custom component driven AI Perception system tailored specifically for implementation in Stealth games. This system uses a combination of four different perception models to evaluate threats based on the new Stimulus model introduced in the update.

Perception Models:

1. Visual Perception: Using a combination of distance, angular, & line trace checks, the Visual Perception enables an AI agent to detect stimuli within it's direct line of sight range. It is capable for sensing Target stimuli as well as certain types of Interest stimuli like Incapacitated & Defunct allies.

2. Aural Perception: The Aural Perception enables an AI agent to perceive & track noises created in the vicinity. It factors in the Loudness of a noise while performing the distance check operation, in order to gauge the relevance of the stimulus. This ensures that louder noises can potentially be heard by an agent, even if it was created outside the default hearing threshold.

3. Intuitive Perception: The Intuitive Perception model is used by AI agents to evaluate stimuli based on their memory, thus negating the need to do any form of range checks. For example, a target that has evaded a Patrol Guard's line of sight can trigger the creation of a Missing Suspect stimulus using it's memory about the target's last seen location, which can then be perceived through this system with very minimal search costs.

4. Motion Perception: Unlike the other perception models, the Motion Perception is used to evaluate stimuli that have not been directly perceived by the agent. An example of this would be a stimulus created by an entity located in close proximity to an agent, but not necessarily within it's direct field of vision. As a result, this system uses a dedicated alert meter to make sure that the agent responds to a perceived stimulus only if it stays within range for certain period of time.


All or any of these systems can be modified or turned off from the owning class, thus facilitating creation of different types of perception behavior among two types of agents like say, a Patrol Guard & a Surveillance Camera.