SmartWalkCoach: An AI Companion for End-to-End Walking Guidance, Motivation, and Reflection Conference: 31st International Conference on Intelligent User Interfaces; March 23–26, 2026; Paphos, Cyprus 31st International Conference on Intelligent User Interfaces (IUI ’26), March 23–26, 2026, Paphos, Cyprus DOI: 10.1145/3742413.3789171 ISBN: 979-8-4007-2281-3/26/04 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Ubiquitous and mobile computing systems and tools CCS: Human-centered computing Empirical studies in HCI CCS: Applied computing Health informatics CCS: Information systems Location based services

Xianzhe Zhang, Mingxuan Hu, Bufan Xue, Erick Purwanto, Thomas Selig, Daniel Yonto · 2026

Read SmartWalkCoach: An AI Companion for End-to-End Walking Guidance, Motivation, and Reflection Conference: 31st International Conference on Intelligent User Interfaces; March 23–26, 2026; Paphos, Cyprus 31st International Conference on Intelligent User Interfaces (IUI ’26), March 23–26, 2026, Paphos, Cyprus DOI: 10.1145/3742413.3789171 ISBN: 979-8-4007-2281-3/26/04 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Ubiquitous and mobile computing systems and tools CCS: Human-centered computing Empirical studies in HCI CCS: Applied computing Health informatics CCS: Information systems Location based services by Xianzhe Zhang, Mingxuan Hu, Bufan Xue, Erick Purwanto, Thomas Selig, Daniel Yonto (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-14628.txt.

Introduction

Zhang and team present SmartWalkCoach, a mobile AI companion that covers a walk from plan to guide to review. Three light agents handle route picks from nearby points of interest, timely prompts that blend facts with encouragement, and short post-walk summaries. In a twelve-person crossover study, adding motivational companion talk is reported to lift positive feelings and user experience. No carryover effects were found, so the gains did not leak from one walk to the next.

Catalog record

CREATORXianzhe Zhang, Mingxuan Hu, Bufan Xue, Erick Purwanto, Thomas Selig, Daniel Yonto
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE69624
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.14628

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

SmartWalkCoach: An AI Companion for End-to-End Walking Guidance, Motivation, and Reflection Conference: 31st International Conference on Intelligent User Interfaces; March 23–26, 2026; Paphos, Cyprus 31st International Conference on Intelligent User Interfaces (IUI ’26), March 23–26, 2026, Paphos, Cyprus DOI: 10.1145/3742413.3789171 ISBN: 979-8-4007-2281-3/26/04 CCS: Human-centered computing Human computer interaction (HCI) CCS: Human-centered computing Ubiquitous and mobile computing systems and tools CCS: Human-centered computing Empirical studies in HCI CCS: Applied computing Health informatics CCS: Information systems Location based services Xianzhe Zhang, Mingxuan Hu, Bufan Xue, Erick Purwanto, Thomas Selig, Daniel Yonto Abstract . We present SmartWalkCoach, a mobile AI companion that supports the full walking journey: from pre-walk planning to in-walk guidance through to post-walk reflection. Addressing a gap between map navigation and motivational coaching, SmartWalkCoach orchestrates three lightweight agents: (1) GeographyAgent for conversational route curation from nearby points of interest and user preferences while delegating pathfinding to map APIs; (2) AccompanyAgent for context-aware, just-in-time prompts that blend informational cues with relational encouragement; and (3) SummaryAgent for concise reflection and next-step planning. This end-to-end, tool-using design aims to lower cognitive load in planning and sustain engagement and motivation during walking through delivering dynamic, cadence-aware interventions. We conducted an in-the-wild, two-period AB/BA crossover study (N=12), where each participant completed two comparable walks with counterbalanced conditions: Information-only versus Information+Motivation. Linear mixed models show that adding motivational, companion-like dialogue significantly improved outcomes: participants reported higher positive feelings and better user experience, with no evidence of carryover. Thematic analysis surfaced two design imperatives for mobile companions: supportive, relational expression and context-aware timing (e.g., avoiding high-load moments, intervening at fatigue/milestones). Our contributions are: (i) an end-to-end, tool-using agent architecture for everyday walking that reduces cognitive load during planning and accompaniment; (ii) a controlled field evaluation linking context-aware motivation to affect and UX gains; and (iii) actionable design guidance on expression, timing, and frequency for mHealth companions. We outline limitations and paths toward multimodal, voice-first companions, with adaptive personalization mechanisms. † † cc-license: by 1. Introduction Walking is one of the foundational elements of healthy living, with its powerful benefits to both physical and mental health being well established ( 94 ; 59 ) . As health awareness expands, numerous applications have been created to promote physical activity. Currently, much of the existing AI/LLM work on walking or exercise support emphasizes in-walk navigation and route direction tasks ( 46 ; 58 ; 50 ; 86 ; 36 ; 14 ) . In parallel, within the health/fitness domain, conversational agents and chatbots have been studied for their ability to promote physical activity, engagement, and behavior change ( 93 ; 64 ; 40 ) . However, several challenges remain inadequately addressed in supporting the complete walking journey. Firstly, a noticeable gap exists between the two aforementioned strands: systems focusing on navigation often lack integrated motivational support, while conversational agents for stimulating physical activity may not be deeply coupled with real-time, context-aware geospatial guidance. Furthermore, many current walking app designs lack universality . Instead, they target specific user groups, such as keen walkers engaging in goal-oriented activities like hiking or racing (see ( 89 ) for a typical example). This focus may limit their effectiveness for everyday walking environments and discourages their uptake by a more diverse range of users which includes occasional walkers. Secondly, in the pre-walk planning phase, users often find it difficult to formulate routes that truly reflect their preferences and needs. There is an inherent information gap between what users intuitively desire and the options presented by map systems. Many users have only partial knowledge of nearby points of interest (POIs) and struggle to express vague but nuanced preferences, such as wanting “a quiet, shaded path with occasional cafés”, and then match those desires with the vast candidate pool. Traditional recommender systems, which rely heavily on rich historical interaction data, tend to suffer in scenarios involving new users or loosely specified preferences (i.e. cold - start conditions). The work ZeroPOIRec is one example of using LLMs to provide zero-shot POI recommendations in such low-data contexts ( 43 ) . More recently, GA - LLM (Geography-Aware LLM) demonstrates how language models augmented with spatial encoding and POI relationship reasoning can improve next-POI prediction performance under sparse data ( 53 ) . Still, applying these advances specifically to walking route planning—where recommendations must jointly satisfy spatial constraints, personal preferences, and real - world accessibility—remains a challenging problem. Finally, another difficulty lies during the walk itself: walkers frequently experience motivational dips, which can lead to incomplete routes ( 4 ) . Current motivation mechanisms within apps are often restricted to passive navigation guidance or rudimentary gamification strategies, such as earning badges for distance covered ( 79 ) . While gamification elements can provide initial excitement, they often fail to sustain intrinsic motivation over the long term ( 2 ) . This approach overlooks the potential for context-aware, real-time companionship, which HCI and agent research suggests is crucial for establishing lasting, trusting relationships between users and computational systems. For instance, relational agents designed for long-term interaction have been shown to be more trusted, respected, and liked compared to task-only agents in month-long trials ( 9 ) . Broader trust models in human-computer interaction emphasize the importance of predictability, consistency, and social cues in building trust over time ( 30 ; 20 ) . Studies in real-time collaboration contexts further indicate that agent behaviors directly influence users’ reliance and confidence in the system ( 19 ) . An ideal walking companion should therefore not only provide timely information but also deliver motivational support tailored to the user’s current physiological and contextual state. This aligns with the principles of Just-In-Time Adaptive Interventions (JITAIs), which advocate for delivering the right type of support at the opportune moment to maximize behavioral impact ( 60 ; 44 ) . We therefore ask the following research questions: RQ1:: How can conversational interfaces for route planning reduce cognitive load and improve user satisfaction in mobile applications? RQ2:: In real-world walking, does an Informative + Motivational agent improve motivation, engagement, and satisfaction compared to an Informative-only agent? RQ3:: How to optimize the expression/frequency/timing of accompanying agents during walking? To address these, we conceived and implemented SmartWalkCoach , an Android-based program that realizes an end-to-end intelligent user interface for walking companionship empowered by AI. SmartWalkCoach brings together three fundamental functional agents working together to oversee the whole walking experience. The GeographyAgent is responsible for conversational route planning by employing geographical API tools alongside solicited user inclinations ( 36 ; 83 ; 43 ) . The AccompanyAgent brings in-situ companionship with contextualized motivational encouragement and map information prompts ( 63 ; 54 ) . Lastly, the SummaryAgent organizes post-walk reflection and consolidation and optional social media sharing ( 80 ) . We then conducted an in-the-wild, walk-along, experimental study to validate the effectiveness of SmartWalkCoach, with 12 users. Each user completed two walks, one under Information+Motivation exposure and the other under Information-only. Researchers followed the users during the walks at a suitable distance in order to take notes on specific behaviors (e.g., stopping to converse with SmartWalkCoach). This in-the-wild observation was conducted as part of a field study approach, allowing the research team to systematically capture users’ natural interactions and behavior

Read the complete text