Emotion-Aware Continuous Stress Monitoring via a Lightweight Conversational Agent for Long-Duration Space Missions

Abstract

Long-duration space missions expose astronauts to sustained psychological stressors including isolation, confinement, high workload, and communication delays. Continuous monitoring of emotional state and stress levels is essential for early detection of psychological deterioration and timely intervention, yet traditional approaches relying on scheduled questionnaires and periodic clinical check-ins provide only discrete snapshots and may miss gradual changes. Conversational AI offers a complementary channel — as astronauts engage with an onboard assistant, their natural language provides continuous signals about mood, stress, and coping. In this paper, we present an emotion-aware stress monitoring framework integrated into MAITRI, an offline conversational assistant for astronaut well-being. The framework automatically infers coarse-grained emotions across six categories (anxious, sad, angry, positive, tired, neutral) and estimates a continuous stress level (1–10 scale) from textual interactions using lightweight, offline-compatible keyword-lexicon methods. Risk assessment classifies each interaction across four severity tiers (low, medium, high, critical) with a non-clinical safety escalation mechanism. All emotion, stress, risk, and rating data are logged in a structured SQLite schema enabling longitudinal trend analysis. We describe the mathematical model underlying emotion scoring and stress estimation, the integrated logging architecture, and an experimental protocol for evaluating framework performance using human-annotated conversation data and self-reported stress measures in a simulated mission context. Proposed evaluation metrics include per-category F1 scores for emotion detection, Pearson correlation and MAE between automatic and self-reported stress, and analysis of relationships between emotional state, stress level, and perceived response helpfulness. This work demonstrates how a lightweight, transparent, privacy-preserving monitoring framework can enable continuous psychological surveillance in resource-constrained environments such as crewed spacecraft.

Introduction

MAITRI, an offline conversational AI system designed for continuous psychological monitoring of astronauts during long-duration space missions.

At its core, the work argues that traditional mental health monitoring in space (interviews, questionnaires, physiological sensors) is too infrequent, delayed, or intrusive to reliably detect gradual psychological changes. Instead, it proposes using natural language conversations with an AI assistant as a continuous, low-burden signal of emotional and stress states.

The system works by analyzing astronaut dialogue in real time using a transparent, lexicon-based model. It detects emotions (like anxious, sad, angry, positive, tired), assigns a stress score (1–10) based on keyword intensity and emotional signals, and classifies risk levels (low to critical) using rule-based thresholds, including safety-critical detection of suicidal ideation. All outputs are logged locally in a structured SQLite database for longitudinal tracking and trend analysis.

The framework is designed with three key components:

1. a mathematical model for emotion detection, stress estimation, and risk classification that is simple, interpretable, and suitable for safety-critical offline use;
2. a logging system that stores emotional and stress data over time for analysis;
3. an experimental evaluation plan to validate accuracy against human annotations and self-reported stress measures.

The background section situates this within prior work in space psychology, affective computing, and language-based mental health detection. It highlights that while modern transformer models are more accurate, they are too resource-heavy for offline spacecraft systems, making lexicon-based methods a practical compromise.

Finally, the paper proposes a 5–7 day simulated study with participants interacting with MAITRI, collecting self-reports, response ratings, and human annotations to evaluate emotion detection accuracy, stress estimation validity, and user trust.

Conclusion

This paper presented an emotion-aware continuous stress monitoring framework integrated into MAITRI, an offline conversational assistant for astronaut psychological well-being. We formalized the mathematical model for lexicon-based emotion scoring, stress estimation, and four-tier risk classification. We described the SQLite logging architecture enabling longitudinal trend analysis, and proposed a comprehensive experimental protocol with human annotation, self-report comparison, and rating correlation analysis for framework evaluation. The framework contributes a practical, privacy-preserving approach to continuous psychological surveillance in resource-constrained environments, using natural conversation as the monitoring medium. By making every interaction an opportunity for non-intrusive assessment, MAITRI\'s monitoring layer supports the early detection of psychological deterioration that is essential for sustained well-being on long-duration space missions. The proposed experimental protocol provides a replicable methodology for evaluating similar frameworks in other isolated, confined, and extreme environments.

References

[1] NASA Human Research Program, \"Behavioral Health and Performance: Evidence Report,\" NASA JSC, Houston, TX, 2019. [2] N. Kanas and D. Manzey, Space Psychology and Psychiatry, 2nd ed. Dordrecht: Springer, 2008. [3] J. Stuster, Bold Endeavors: Lessons from Polar and Space Exploration. Naval Institute Press, 2010. [4] L. Landon, J. Slack, and L. B. Barrett, \"Teamwork and collaboration in long-duration space missions,\" American Psychologist, vol. 73, no. 4, pp. 563–575, 2018. [5] A. Miner, A. Milstein, and J. Hancock, \"Talking to machines about personal mental health problems,\" JAMA, vol. 318, no. 13, pp. 1217–1218, 2017. [6] S. Bane, \"MAITRI: An offline LLM-based assistant for astronaut psychological well-being,\" M.Tech Project, GHRCEM, Pune, 2025–26. [7] E. A. Boyle et al., \"Behavioral health and performance in isolated and confined environments,\" Aerospace Medicine and Human Performance, vol. 90, no. 5, 2019. [8] J. W. Pennebaker, R. J. Booth, and M. E. Francis, \"LIWC2015 Operator\'s Manual,\" Pennebaker Conglomerates, Austin, TX, 2015. [9] Y. R. Tausczik and J. W. Pennebaker, \"The psychological meaning of words: LIWC and computerized text analysis,\" Journal of Language and Social Psychology, vol. 29, no. 1, pp. 24–54, 2010. [10] S. M. Mohammad and P. D. Turney, \"Crowdsourcing a word-emotion association lexicon,\" Computational Intelligence, vol. 29, no. 3, pp. 436–465, 2013. [11] C. J. Hutto and E. Gilbert, \"VADER: A parsimonious rule-based model for sentiment analysis,\" ICWSM, 2014. [12] Y. Liu et al., \"RoBERTa: A robustly optimized BERT pretraining approach,\" arXiv:1907.11692, 2019. [13] D. Demszky et al., \"GoEmotions: A dataset of fine-grained emotions,\" ACL, 2020. [14] N. Bao et al., \"Stress detection using social media text,\" IEEE Access, 2021. [15] D. Guntuku et al., \"Detecting stress in social media,\" EMNLP, 2017. [16] M. Abd-Alrazaq et al., \"Systematic review of AI chatbot effectiveness for mental health,\" JMIR Mental Health, 2021. [17] S. Cohen, T. Kamarck, and R. Mermelstein, \"A global measure of perceived stress,\" J. Health and Social Behavior, vol. 24, pp. 385–396, 1983. [18] K. K. Fitzpatrick, A. Darcy, and M. Vierhile, \"Delivering CBT using a conversational agent (Woebot),\" JMIR Mental Health, 2017. [19] C. D. Spielberger, State-Trait Anxiety Inventory. Consulting Psychologists Press, 1983. [20] K. Kroenke, R. Spitzer, and J. Williams, \"The Patient Health Questionnaire-2,\" Medical Care, vol. 41, pp. 1284–1292, 2003. [21] J. Brooke, \"SUS: A quick and dirty usability scale,\" Usability Evaluation in Industry, pp. 189–194, 1996.

Copyright

Copyright © 2026 Sakshi Bane, Dr. Sarita Patil. 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.