Featured Application
AIoTutor can be deployed in embedded IoT laboratory courses to provide teacher-vetted, tiered troubleshooting guidance on commodity hardware; instructors archive high-frequency faults into a reviewed knowledge base, reducing repetitive one-on-one troubleshooting in large-cohort laboratories.
Abstract
Embedded Internet of Things (IoT) laboratory courses face two persistent challenges: novice students bear excessive extraneous cognitive load in cross-layer hardware–software fault diagnosis, and teachers’ tacit troubleshooting pedagogical content knowledge (PCK) cannot be systematically preserved. This paper presents AIoTutor, a four-layer scaffolded generative artificial intelligence tutoring intervention grounded in cognitive load theory, the Technological Pedagogical Content Knowledge framework, and self-regulated learning theory, centered on a mandatory teacher knowledge-review pipeline. A four-group sequential ablation quasi-experiment with 96 engineering undergraduates compared the full intervention against raw large language model support, unreviewed knowledge base support, and traditional instruction. Compared with traditional instruction, the full intervention was associated with higher post-test conceptual mastery (78.42 vs. 66.96, p < 0.001), 42.9% lower perceived cognitive load (4.61 vs. 8.08), larger self-regulated learning gains (3.54 vs. 0.83), and an 82.7% reduction in repetitive teacher troubleshooting workload (14 vs. 81 interventions). A delayed transfer test with novel faults showed the highest independent fault-location rate under the full intervention (58.3%, 52.1%, 39.6%, and 29.2% for G1, G3, G2, and G4); because transfer observations are nested within students, these results are treated as exploratory, with only the full-intervention versus control contrast reaching significance (mixed-effects model, p = 0.028). Blind-rated answer-quality analysis showed that teacher revision primarily improved pedagogical rather than technical dimensions, a pattern interpreted as preliminary associative evidence given the sequential design.