Individuals with high-functioning autism and Asperger Syndrome find everyday social interactions very challenging (Rapin & Tuchman, 2008). Sufferers often feel more comfortable interacting with computers and robots than their peers (Putnam & Chong, 2008). A computer program providing a virtual human tutor that teaches social skills can capitalise on this preference for technology, while working towards improved social competence. Such an approach is expected to provide a motivating, non-judgemental environment in which to develop social skills, while offering a tool that can be used independently by the individual, relieving pressure from carers, educators and therapists working with the learner. While little existing research has focussed on using embodied conversational agents (ECAs) for teaching children with autism, what has been done has led to promising outcomes (see Bosseler & Massaro, 2003; Tartaro & Cassell, 2006). In developing an ECA for improving social competence in children with autism, the particular needs of these individuals must be taken into account. Very importantly, this means that the tutor must present content in a way that suits the learning style and sensory needs of the learner, provide assessment tasks that are both appropriate and relevant and interpret the results of these assessments in a meaningful manner.

Embodied Conversational Agents (ECAs) are autonomous, animated virtual agents that interact with users in real-time dialogue, recognising and performing both speech and gesture, much akin to human-to-human interaction (Cassell, 2001). To implement a virtual embodied conversational agent requires use of a range of cutting edge technologies and significant computational power. Thus, most agents referred to as ECAs are not truly embodied in the sense of the given definition, with many relying on heuristics to give the appearance of embodiment while being computationally efficient enough to run on standard personal computers. In this chapter, we use the term 'virtual agent' to refer to a virtual entity with some autonomous decision making capabilities. Virtual agents can be broadly categorised as being 'authorable' or 'autonomous', with authorable agents requiring someone to observe the user and select from a control panel actions for the agent to take, much like a virtual puppet, and autonomous agents being self-contained and usable without such outside interaction.

2.1 Virtual Tutors and Why They Can Benefit Children with Autism

ECAs are used in a range of applications. When used in an educational context, they are often referred to as virtual tutors or pedagogical agents. There are a multitude of reasons why virtual tutors are well suited to use with individuals with autism, not least of which is the affinity that many report having with computers and technology (Putnam & Chong, 2008). A technology based intervention for improving social competence appeals to children with autism and allows them to progress through material at their own pace rather than being subjected to the constraints of a classroom. It is openly acknowledged that nothing should aim to replace interaction with peers and others when learning about social interaction. However, an independent learning tool, such as one incorporating a virtual tutor, can provide a valuable first step in developing these complex social skills.