From co-designed personas to in-the-wild deployment: a therapist-operable social robot platform for pediatric speech-language therapy

Socially Assistive Robots (SARs) have been extensively studied in pediatric therapy for neurodivergent children, yet their use for speech-language intervention in routine clinical practice remains underexplored. We present a therapist-operable, robot-assisted therapy platform consisting of a physical social robot and a web-based control dashboard, co-developed with Speech-Language Pathologists (SLPs) through a multi-year co-design process to support independent robot use during regular sessions without on-site technical support. We report a two-stage system evaluation: (1) iterative co-design and prototype validation that translates personas into implementable requirements and therapy activities (Study 1); and (2) an in-the-wild deployment with one SLP and six children diagnosed with receptive and expressive language delay, consisting of seven weekly sessions per child over an 11-week calendar period (Study 2). No researchers were present on site during the sessions. Results show that the robot effectively supported attention re-engagement, emotional regulation during meltdowns, and motivation via reward-like interactions. However, sustained use revealed key challenges for long-term viability: developmental mismatch between prepared activities and individual functional levels, limited adaptability and content scalability for repeated sessions, and uncertainty about acceptable physical interaction boundaries. We synthesize these results into traceable design-to-deployment implications and generalizable principles for therapist-operated SAR systems.