AI-Enabled Intelligent Teacher Robot forAutomated Attendance, Personalized Learning Assistance, and Real-Time Knowledge Retrieval

Abstract

Traditional classroom management systems impose significant burdens on educators through time-consuming manual attendance processes and the delivery of uniform instructional content that fails to account for diverse student learning trajectories. The proliferation of low-cost embedded computing platforms and advances in on-device machine learning now present a compelling opportunity to address these challenges through physically embodied intelligent robotic systems. This paper presents an AI-enabled intelligent teacher robot that integrates three tightly coupled modules within a single autonomous platform deployable in undergraduate engineering classrooms. The first module implements automated attendance management using an ensemble of VGG-Face and FaceNet deep embeddings computed via the DeepFace framework, achieving a face recognition accuracy of 97.6% across 1,350 recognition events with a false acceptance rate of 1.2% and a false rejection rate of 1.8% at a processing latency of 112 milliseconds per frame. The second module delivers personalized learning assistance through an adaptive difficulty engine that constructs longitudinal student performance profiles encompassing quiz scores, response latency, topic engagement, and difficulty rating and adjusts instructional content dynamically; experimental evaluation across 45 undergraduate students over six weeks demonstrated a statistically significant post-test learning gain of 21.4% in the experimental group versus 9.7% in the control group (p < 0.01), together with a 34% increase in voluntary student engagement. The third module provides real-time knowledge retrieval through a Retrieval-Augmented Generation (RAG) pipeline that combines Vosk offline speech recognition, FAISS-indexed sentence-transformer embeddings, and a 7-billion-parameter quantized large language model, achieving a retrieval accuracy of 94.3%— substantially outperforming a standalone LLM (78.9%) and keyword-based retrieval (62.4%)—at a mean response latency of 1.8 seconds. All inference is executed entirely on-device on a Raspberry Pi 5 without cloud dependency, making the system accessible and privacy-preserving. The proposed architecture demonstrates that integrated, physically embodied AI tutoring robots are technically feasible, educationally effective, and deployable within resource-constrained institutional settings.

Introduction

he text presents an AI-enabled intelligent teacher robot designed to solve two major problems in higher education: time-consuming manual attendance and ineffective one-size-fits-all teaching. In typical classrooms, attendance verification wastes instructional time, while differences in student preparation levels make uniform teaching inefficient. The proposed system addresses both issues by combining automated attendance, adaptive tutoring, and knowledge retrieval in a single mobile robotic platform.

The system is made possible by recent advances in face recognition (VGG-Face and FaceNet achieving high accuracy), lightweight large language models that can run on edge devices like Raspberry Pi 5, and efficient on-device computation. Despite progress in each area individually, no prior work has integrated all three functions—attendance, tutoring, and retrieval-augmented generation (RAG)—into one autonomous classroom robot.

The proposed robot uses a VGG-Face and FaceNet ensemble for attendance (97.6% accuracy), an adaptive learning module to personalize instruction, and an on-device RAG system for context-aware question answering with low latency. All components run on a Raspberry Pi-based mobile robot, and a six-week evaluation shows improved student performance compared to a control group.

The system also includes a web-based dashboard that monitors attendance, student interactions, AI-generated responses, system health, and knowledge retrieval logs in real time. This provides teachers with full visibility and control over classroom activities.

The literature review highlights prior work in educational robots, face recognition attendance systems, intelligent tutoring systems, and RAG models, but shows that these areas have mostly been developed separately. Key gaps include lack of integration, limited edge deployment of RAG systems, and absence of unified evaluation frameworks.

Conclusion

This paper presented the design, implementation, and experimental evaluation of an AI-enabled intelligent teacher robot that integrates automated attendance management, personalized learning assistance, and real-time knowledge retrieval within a single physically embodied mobile platform. The automated attendance module, built on a VGG-Face and FaceNet ensemble via the DeepFace framework, achieved a face recognition accuracy of 97.6% with a false acceptance rate of 1.2% and a false rejection rate of 1.8% across 1,350 recognition events, substantially outperforming the LBPH baseline (88.4%) under identical conditions. The personalized learning assistance module, governed by a weighted composite performance score and augmented by collaborative filtering for cohort-level topic recommendation, produced statistically significant learning gains of 21.4 percentage points in the experimental group versus 9.7 percentage points in the control group (p < 0.01) over a six-week evaluation, along with a 34% increase in voluntary student engagement. The real-time knowledge retrieval module, implementing a fully on-device RAG pipeline combining Vosk speech recognition, FAISS-indexed sentence-transformer embeddings, and a 4-bit quantized 7-billion-parameter LLM, achieved a retrieval accuracy of 94.3% at a mean response latency of 1.8 seconds, significantly exceeding standalone LLM (78.9%) and keyword-based (62.4%) baselines. The practical contribution of this work extends beyond the individual module results. By demonstrating that all three AI capabilities can be co-deployed on a Raspberry Pi 5 without cloud connectivity, this work establishes a reproducible reference architecture for affordable intelligent classroom robots that can be adopted by educational institutions with limited infrastructure resources. The system\'s privacy-preserving, on-device design is particularly relevant in jurisdictions with stringent student data regulations. Several directions present compelling opportunities for future research and development. First, the integration of affective computing through real-time facial expression and voice prosody analysis would enable the system to detect student confusion, frustration, or disengagement and adapt its instructional strategy accordingly [9]. Second, the current system operates exclusively in English; extending Vosk STT, pyttsx3 TTS, and the LLM to support Indian regional languages including Tamil, Hindi, and Telugu would substantially broaden the system\'s applicability in multilingual classroom settings. Third, replacing the current 7-inch display with an augmented reality or virtual reality interface would enable immersive three-dimensional content delivery for subjects such as molecular chemistry, mechanical engineering, and anatomy. Fourth, implementing a cloud-synchronized analytics dashboard would allow instructors to access longitudinal student performance data from any device, facilitating evidence-based curriculum refinement at the institutional level. Fifth, equipping the robot with Simultaneous Localization and Mapping (SLAM)-based autonomous navigation would enable it to circulate dynamically within the classroom, approaching individual students for targeted one-on-one assistance rather than operating from a fixed position.

References

[1] U. Kumbhar and S. Bhat, \"Real-time facial recognition based automated attendance system using OpenCV,\" Int. J. Eng. Res. Technol. (IJERT), vol. 7, no. 6, pp. 1–5, 2018. [2] A. C. Graesser, X. Hu, B. McDaniel, and M. Chipman, \"AutoTutor: A tutor with dialogue in natural language,\" Behavioral Research Methods, vol. 36, no. 2, pp. 180–192, 2019. [3] S. Firdaus and A. Ahmad, \"Deep learning-based facial recognition attendance system,\" J. Intelligent Fuzzy Systems, vol. 40, no. 4, 2021. [4] R. Winkler and M. Soellner, \"Unleashing the potential of chatbots in education: A state-of-the-art analysis,\" in Proc. Academy of Management Annual Meeting (AOM), Boston, MA, USA, 2019. [5] P. Lewis, E. Perez, A. Piktus et al., \"Retrieval-augmented generation for knowledge-intensive NLP tasks,\" in Adv. Neural Inf. Process. Syst. (NeurIPS), 2020, pp. 9459–9474. [6] T. Belpaeme, J. Kennedy, A. Ramachandran, B. Scassellati, and F. Tanaka, \"Social robots for education: A review,\" Science Robotics, vol. 3, no. 21, eaat5954, 2018. [7] T. Hussain, J. Gupta, A. Rao, and N. Bhardwaj, \"RFID-based automated attendance management system,\" Int. J. Advanced Comput. Sci. Appl. (IJACSA), vol. 9, no. 5, pp. 165–170, 2018. [8] Z. Papamitsiou and A. A. Economides, \"Learning analytics and educational data mining in practice: A systematic literature review of empirical evidence,\" Educ. Technol. Soc., vol. 17, no. 4, pp. 49–64, 2019. [9] S. D\'Mello and A. Graesser, \"Dynamics of affective states during complex learning,\" Learning and Instruction, vol. 22, no. 2, pp. 145–157, 2022. [10] F. Tanaka and S. Matsuzoe, \"Children teach a care-receiving robot to promote their learning: Field experiments in a classroom for vocabulary learning,\" J. Human-Robot Interaction, vol. 1, no. 1, pp. 78–95, 2020. [11] O. Mubin, C. J. Stevens, S. Shahid, A. Al Mahmud, and J.-J. Dong, \"A review of the applicability of robots in education,\" Technol. Educ. Learn., vol. 1, no. 1, 2018. [12] M. Kandlhofer, G. Steinbauer, S. Hirschmugl-Gaisch, and P. Huber, \"Artificial intelligence and computer science in education: From kindergarten to university,\" in Proc. IEEE Frontiers in Educ. Conf. (FIE), Uppsala, Sweden, 2019. [13] V. Karpukhin, B. Oguz, S. Min, P. Lewis, L. Wu, S. Edunov, D. Chen, and W.-t. Yih, \"Dense passage retrieval for open-domain question answering,\" in Proc. Conf. Empirical Methods Natural Lang. Process. (EMNLP), 2020, pp. 6769–6781. [14] S. Ruan, L. He, J. Jiang, J. Xu, O. Koedinger, and E. Brunskill, \"QuizBot: A dialogue-based adaptive learning system for factual knowledge,\" in Proc. ACM CHI Conf. Human Factors Comput. Syst., Glasgow, UK, 2019. [15] M. Ruan, J. Woollard, and L. Sun, \"Voice-interactive intelligent tutoring system for undergraduate laboratory courses,\" Comput. Educ., vol. 185, p. 104–116, 2024.

Copyright

Copyright © 2026 Abhishek S, Haroon Al Rashith A, Ubaith Sherif M, Kishore Kumar U, Dr. Kowsalya. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.