Virtual therapists and AI-driven systems are emerging as promising tools in motor rehabilitation, offering opportunities to enhance therapeutic outcomes, increase accessibility, and support clinician workload. We synthesize a systematic literature review and recent participatory design research involving therapists to critically examine virtual therapists' current and future role in immersive virtual environments.

Our review found that virtual therapists are predominantly human-like avatars that demonstrate movement tasks for patient imitation, often supported by visual, haptic, or unidirectional audio feedback. User studies indicate high patient motivation and therapist acceptance, but current systems are limited by the absence of bidirectional communication, constrained personalisation, and limited attention to therapeutic alliance.

Our recent co-design workshops with therapists highlighted three key design principles for next-generation virtual systems: preserving human connection, fostering trust through transparent interaction, and enabling individualised therapy through adaptable engagement. Together, these insights suggest that AI-powered virtual therapists should serve as supportive, responsive guides, enhancing, rather than replacing, the human therapeutic relationship.

We argue that emerging AI capabilities such as natural language processing, adaptive learning, and real-time progress monitoring offer compelling avenues for innovation. However, successful implementation requires ethically grounded, user-centred development that considers the cognitive and emotional needs of diverse patient populations. Participatory approaches involving both patients and clinicians will be essential in shaping virtual systems that are accessible, effective, and therapeutically meaningful.

Hence, we advocate for interdisciplinary collaboration, clinical validation, and thoughtful integration of AI to ensure virtual therapists become a trusted and impactful component of future motor rehabilitation practice.

• https://programme.conventus.de/en/dgnr-2...