An Investigation Into the Use of Synthetic Vision for NPC’s / Agents in Computer Games

Documents

mick
  • An Investigation Into the Use of Synthetic Vision for NPC’s / Agents in Computer Games Paper by Sebastian Enrique Presented by Adam Karkkainen
  • Table of Contents Introduction Problem Statement Synthetic Vision Model Brain Module Known Problems Extended Behavior with Dynamic Reactions Conclusions Future Work
  • Introduction Today, 3D computer games take information directly from the internal database for the non-player characters (NPC’s). The use of synthetic vision together with complex brain modules could improve gameplay and make for better and more realistic NPC’s.
  • How computer vision issues are addressed Depth perception – pixel depth is used to determine the actual position of objects in the agent’s field of view by inverting the modeling and projection transform. Object recognition – object function or identity part of the system. Motion determination – Code the motion of object pixels into the synthetic vision viewport.
  • Problem Statement Create a visual system of a virtual character who lives in a 3D virtual world. Uses the scene rendered from the character’s point of view.
  • Synthetic Vision Model The AI system should eventually be able to do the following with the model: Obstacle avoidance. Low-level navigation. Fast object recognition. Fast dynamic object detection.
  • Synthetic Vision Model (cont) To do this, the synthetic vision approach uses two viewports. static information viewport. dynamic information viewport.
  • Static Information Viewport Floor : Any polygon from the level geometry with Z normal normalized component greater or equal than 0.8. Ceiling : Any polygon from the level geometry with Z normal normalized component less or equal than -0.8. Wall : Every polygon from the level geometry that does not fit in any of the two previous cases.
  • Objects and Color IDs Mapping Table with Level Geometry
  • Dynamic Information Viewport Represents instantaneous movement information of the objects being seen in the static viewport. Fully static object would have a color of (0.0, 0.5, 1.0) Moving objects will have different colors based on movement direction and velocity
  • Brain Module Takes the synthetic vision module as input, processes it, and acts after it Very basic AI, only enough to do proof of concept for synthetic vision If health or ammo drops below a certain threshold, search for pickups Health and ammo drop periodically to facilitate testing
  • NPC Behavior Walk around Looking for Health Looking for Ammo Looking for any Pick-up Looking Quickly for Weapon Looking Quickly for Health
  • Walking Around Algorithm If a central free way exists, it is taken. Try right most free way of the left half of the viewport. Try left most free way of the right half of the viewport. Fails if a thin corridor or a blocking wall is detected. no central free way In the viewport bottom’s closest row that contains at least one pixel different than floor, the two pixels closest to the left and right side of the searching rectangle are at the same distance to the left and right sides. Randomly turn to the left or right and try again
  • Example Free Way Tests
  • Looking for Pickups Algorithm If a % of current path has not been walked, algorithm returns. Create a list called objectlist of the wanted class power-ups seen in the static viewport at that moment. If no wanted class power-ups are seen, Walk Around. If at least one is found, get the closest from objectlist, p. Get floor pixel located in straight vertical line under p and with approximately the same depth value, d. If d does not exist, use Walk Around. If it exists, set d as the new destination coordinates.
  • Known Problems Higher Floor Problem Perspective Problem Looking-For Problem
  • Higher Floor Problem Floor heights are not checked A box that is too high to climb onto but short enough so the upper face is seen Static Viewport represents it as a floor
  • Perspective Problem Free way algorithm has difficulty with corridors due to the trapezoidal shape Also has difficulty with columns close to each other
  • Looking-For Problem When looking for an object, the Actor doesn’t take into account obstacles or the Actor’s width
  • Extended Behavior with Dynamic Reactions Reactive and Rule-based AI to use the information provided in the dynamic viewport Three States for the Actor Intercept Avoid Don’t Worry
  • The Three States Intercept At least one enemy is approaching Actor has health or weapon above upper thresholds Avoid At least one enemy is approaching Actor has health or weapon below the upper thresholds Don’t Worry No enemies or they are all going away from the actor
  • Conclusions More manageable information than pure vision systems. Avoids unrestricted access to the game database. Simple rule-based AI model made to demonstrate the model in a computer game
  • Future Work Enhancements to Synthetic Vision Infrared vision, heat sensing (like the Predator) Lighting effects Adding noise to the vision representation Enhancements to the Brain Memory Learning Interaction between Agents Personality, based of previous three enhancements
  • Questions? Paper at: http://www1.cs.columbia.edu/~senrique/files/thesis_english.pdf
Please download to view
1
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Description
An Investigation Into the Use of Synthetic Vision for NPC’s / Agents in Computer Games. Paper by Sebastian Enrique Presented by Adam Karkkainen. Table of Contents. Introduction Problem Statement Synthetic Vision Model Brain Module Known Problems Extended Behavior with Dynamic Reactions - PowerPoint PPT Presentation
Text
  • An Investigation Into the Use of Synthetic Vision for NPC’s / Agents in Computer Games Paper by Sebastian Enrique Presented by Adam Karkkainen
  • Table of Contents Introduction Problem Statement Synthetic Vision Model Brain Module Known Problems Extended Behavior with Dynamic Reactions Conclusions Future Work
  • Introduction Today, 3D computer games take information directly from the internal database for the non-player characters (NPC’s). The use of synthetic vision together with complex brain modules could improve gameplay and make for better and more realistic NPC’s.
  • How computer vision issues are addressed Depth perception – pixel depth is used to determine the actual position of objects in the agent’s field of view by inverting the modeling and projection transform. Object recognition – object function or identity part of the system. Motion determination – Code the motion of object pixels into the synthetic vision viewport.
  • Problem Statement Create a visual system of a virtual character who lives in a 3D virtual world. Uses the scene rendered from the character’s point of view.
  • Synthetic Vision Model The AI system should eventually be able to do the following with the model: Obstacle avoidance. Low-level navigation. Fast object recognition. Fast dynamic object detection.
  • Synthetic Vision Model (cont) To do this, the synthetic vision approach uses two viewports. static information viewport. dynamic information viewport.
  • Static Information Viewport Floor : Any polygon from the level geometry with Z normal normalized component greater or equal than 0.8. Ceiling : Any polygon from the level geometry with Z normal normalized component less or equal than -0.8. Wall : Every polygon from the level geometry that does not fit in any of the two previous cases.
  • Objects and Color IDs Mapping Table with Level Geometry
  • Dynamic Information Viewport Represents instantaneous movement information of the objects being seen in the static viewport. Fully static object would have a color of (0.0, 0.5, 1.0) Moving objects will have different colors based on movement direction and velocity
  • Brain Module Takes the synthetic vision module as input, processes it, and acts after it Very basic AI, only enough to do proof of concept for synthetic vision If health or ammo drops below a certain threshold, search for pickups Health and ammo drop periodically to facilitate testing
  • NPC Behavior Walk around Looking for Health Looking for Ammo Looking for any Pick-up Looking Quickly for Weapon Looking Quickly for Health
  • Walking Around Algorithm If a central free way exists, it is taken. Try right most free way of the left half of the viewport. Try left most free way of the right half of the viewport. Fails if a thin corridor or a blocking wall is detected. no central free way In the viewport bottom’s closest row that contains at least one pixel different than floor, the two pixels closest to the left and right side of the searching rectangle are at the same distance to the left and right sides. Randomly turn to the left or right and try again
  • Example Free Way Tests
  • Looking for Pickups Algorithm If a % of current path has not been walked, algorithm returns. Create a list called objectlist of the wanted class power-ups seen in the static viewport at that moment. If no wanted class power-ups are seen, Walk Around. If at least one is found, get the closest from objectlist, p. Get floor pixel located in straight vertical line under p and with approximately the same depth value, d. If d does not exist, use Walk Around. If it exists, set d as the new destination coordinates.
  • Known Problems Higher Floor Problem Perspective Problem Looking-For Problem
  • Higher Floor Problem Floor heights are not checked A box that is too high to climb onto but short enough so the upper face is seen Static Viewport represents it as a floor
  • Perspective Problem Free way algorithm has difficulty with corridors due to the trapezoidal shape Also has difficulty with columns close to each other
  • Looking-For Problem When looking for an object, the Actor doesn’t take into account obstacles or the Actor’s width
  • Extended Behavior with Dynamic Reactions Reactive and Rule-based AI to use the information provided in the dynamic viewport Three States for the Actor Intercept Avoid Don’t Worry
  • The Three States Intercept At least one enemy is approaching Actor has health or weapon above upper thresholds Avoid At least one enemy is approaching Actor has health or weapon below the upper thresholds Don’t Worry No enemies or they are all going away from the actor
  • Conclusions More manageable information than pure vision systems. Avoids unrestricted access to the game database. Simple rule-based AI model made to demonstrate the model in a computer game
  • Future Work Enhancements to Synthetic Vision Infrared vision, heat sensing (like the Predator) Lighting effects Adding noise to the vision representation Enhancements to the Brain Memory Learning Interaction between Agents Personality, based of previous three enhancements
  • Questions? Paper at: http://www1.cs.columbia.edu/~senrique/files/thesis_english.pdf
Comments
Top