Towards Truly Autonomous Synthetic Characters with the Sigma Cognitive Architecture

Realism is required not only for how synthetic characters look but also for how they behave. Many applications, such as simulations, virtual worlds, and video games, require computational models of intelligence that generate realistic and credible behavior for the participating synthetic characters. Sigma (S) is being built as a computational model of general intelligence with a long-term goal of understanding and replicating the architecture of the mind; i.e., the fixed structure underlying intelligent behavior. Sigma leverages probabilistic graphical models towards a uniform grand unification of not only traditional cognitive capabilities but also key non-cognitive aspects, creating unique opportunities for the construction of new kinds of non-modular behavioral models. These ambitions strive for the complete control of synthetic characters that behave as humanly as possible. In this paper, Sigma is introduced along with two disparate proof-of-concept virtual humans – one conversational and the other a pair of ambulatory agents – that demonstrate its diverse capabilities.

Learned Behavior

A key challenge in programming virtual environments is to produce virtual characters that are autonomous and capable of action selections that appear believable. In this chapter, motivations are used as a basis for learning using reinforcements. With motives driving the decisions of characters, their actions will appear less structured and repetitious, and more human in nature. This will also allow developers to easily create virtual characters with specific motivations, based mostly on their narrative purposes or roles in the virtual world. With minimum and maximum desirable motive values, the characters use reinforcement learning to drive action selection to maximize their rewards across all motives. Experimental results show that a character can learn to satisfy as many as four motives, even with significantly delayed rewards, and motive changes that are caused by other characters in the world. While the actions tested are simple in nature, they show the potential of a more complicated motivation driven reinforcement learning system. The developer need only define a character's motivations, and the character will learn to act realistically over time in the virtual environment.

A Universal Architecture for Migrating Cognitive Agents

In this chapter, the characteristics of a cognitive architecture that can migrate among various embodiments are discussed and the feasibility of designing such architecture is investigated. The migration refers to the ability of an agent to transfer its internal state among different embodiments without altering its underlying cognitive processes. Designing such architecture will address both weak and strong aspects of AI. The authors propose a Universal Migrating Cognitive Agent (UMCA) inspired by onboard autonomous frameworks utilized in interplanetary missions in which the embodiment can be tailored by defining a set of possible actions and perceptions associated with the new body. The proposed architecture is then evaluated within a few virtual environments to analyze the consistency between its deliberative and reactive behaviors. Finally, UMCA is tailored to automatically create computer animations using a natural language interface.

The Contemporary Craft of Creating Characters Meets Today's Cognitive Architectures

What does the contemporary craft of character design (by human authors), which is beyond the reach of foreseeable AI, and which isn't powered by any stunning, speculative, AI-infused technology (immersive or otherwise), but is instead aided by tried-and-true “AI-less” software tools and immemorial techniques that are still routinely taught today, imply with respect to today's computational cognitive architectures? This chapter narrows the scope of this large question, and argues that at present, perhaps only the cognitive architecture CLARION can represent and reason over knowledge at a level of logical expressivity sufficient to capture such characters, along with the robust modeling implied by contemporary story and character design.

Personality-Based Cognitive Design of Characters in Virtual Environments

Personality-based cognitive architectures should yield consistent patterns of behaviour through personality traits that have a modulatory influence at different levels: These factors affect, on the one hand, high-level components such as ‘emotional reactions' and ‘coping behaviour', and on the other hand, low-level parameters such as the ‘speed of movements and repetition of gestures. In our hybrid cognitive architecture, a deliberative reasoning about the world (e.g. strategies and goals of the 3D character) is combined with dynamic real-time response to the environment's changes and sensors' input (e.g. emotional changes). Hybrid system copes dynamically with changes in the environment, and is complicated enough to have reasoning abilities. Designing a cognitive architecture that gives the impression of personality to 3D agents can be a tremendous help making 3D characters more engaging and successful in interactions with humans.

Integrating ACT-R Cognitive Models with the Unity Game Engine

The main aim of the chapter is to describe how cognitive models, developed using the ACT-R cognitive architecture, can be integrated with the Unity game engine in order to support the intelligent control of virtual characters in both 2D and 3D virtual environments. ACT-R is a cognitive architecture that has been widely used to model various aspects of human cognition, such as learning, memory, problem-solving, reasoning and so on. Unity, on the other hand, is a very popular game engine that can be used to develop 2D and 3D environments for both game and non-game purposes. The ability to integrate ACT-R cognitive models with the Unity game engine thus supports the effort to create virtual characters that incorporate at least some of the capabilities and constraints of the human cognitive system.

Game AGI beyond Characters

Artificial General Intelligence has traditionally used games as a testbed to develop domain-agnostic game playing techniques. Yet games are about more than winning. This chapter reviews recent efforts that have broadened the ways Artificial Intelligence (AI) is used in games, covering: modeling and managing player experiences, creating novel game structures based in interacting with AI, and enabling AI agents to make games. Many of the techniques used to address these challenges have been ad hoc approaches to solving specific problems. This chapter discusses open challenges in each of these areas and the potential for cognitive architectures to provide unified techniques that address these challenges.

Virtual Soar-Agent Implementations

This chapter provides a brief overview of those virtual agent implementations directly inspired by the cognitive architecture: Soar. This chapter will take a qualitative approach to discussing examples of virtual Soar-agents. Finally, this chapter will speculate on the future of Soar virtual characters. The goals of this chapter are sixfold. The first goal is to explain why cognitive architectures are becoming increasingly important to virtual agent design(s). The second goal is to convey why this chapter focuses exclusively on virtual agents that utilize the Soar architecture. The third goal is to explore some of Soar's technical details. The fourth goal is to showcase a few diverse examples where Soar is beginning to have a design impact on virtual agents. The fifth goal addresses Soar's limitations – when applied to agent design in virtual environments. The final goal speculates on ways Soar can be expanded for virtual agent design(s) in the future.

A Graphical Tool for the Creation of Behaviors in Virtual Worlds

Typically, the creation of AI-based behaviors for Non-Playing Characters (NPC) in computer games is carried out by people with specialized skills in the broad areas of design and programming. This book chapter presents Second Mind (SM), a digital solution that makes it possible for novice users to successfully author behaviors. In SM, authors can use a graphical interface to define the behaviors of virtual characters in response to interactions with players. Authors can easily endow NPCs with behavior capabilities that make them act in different possible roles such as e.g. shopkeepers, museum hosts, etc. A series of user tests with human subjects to evaluate the behavior authoring process shows that Second Mind is easy to understand and simplifies the process of behavior production.

On Vision-Based Human-Centric Virtual Character Design

Enabling cognition in a Virtual Character (VC) may be an exciting endeavor for its designer and for the character. A typical VC interacts primarily with its virtual world, but given some sensory capabilities (vision or hearing), it would be expected to explore some of the real world and interact with the intelligent beings there. Thus a virtual character should be equipped with some algorithms to localize and track humans (e.g. via 2D or 3D models), recognize them (e.g. by their faces) and communicate with them. Such perceptual capabilities prompt a sophisticated Cognitive Architecture (CA) to be integrated into the design of a virtual character, which should enable a VC to learn from intelligent beings and reason like one. To seem natural, this CA needs to be fairly seamless, reliable and adaptive. Hence a vision-based human-centric approach to the VC design is explored here.