Despite the widespread adoption of robots across various industries, service-oriented sectors such as hospitality continue to face significant limitations in deploying robots due to the inherently unstructured nature of service tasks. Customers still tend to prefer interactions with human employees, while exceptional situations that fall outside predefined procedures frequently arise in practice. Therefore, rather than pursuing the complete replacement of human employees, there is a growing need to establish a systematic and interdependent human–robot cooperation framework. This study proposes an optimal operational system for interdependent human–robot cooperation in restaurant environments and redefines robots not merely as substitutes for human labor but as complementary partners. This study developed a multi-objective optimization model that simultaneously minimizes total operating costs, including robot subscription fees and employee labor costs, as well as employees’ physical workload, while satisfying predefined customer satisfaction targets. In the proposed framework, repetitive and physically demanding transportation tasks are assigned to robots, whereas personalized face-to-face services are performed by human employees. Furthermore, realistic operational constraints are mathematically incorporated, including employee break requirements mandated by labor regulations, robot battery charging requirements, and the negative impact of accumulated employee fatigue on service quality. To address operational problems of varying scales, this study implements a staged solution approach that combines mathematical optimization techniques for obtaining baseline optimal solutions, metaheuristic algorithms for large-scale operational problems, and reinforcement learning methods to capture uncertainty in real-time environments. This study presents a leading framework at the intersection of service operations management and ergonomics and provides practical insights for sustainable technology integration in service industries.

This study proposes a reinforcement learning-based collaborative operation model between hotel service robots and human employees to address the dynamic task allocation problem in hotel service environments. Hotel service operations involve various customer requests that occur unpredictably over time, while the availability of robots and employees continuously changes according to operational conditions. Consequently, conventional static optimization approaches have limitations in effectively reflecting the uncertainty and real-time characteristics of actual hotel operation. To represent a realistic service environment, a virtual hotel consisting of 15 floors and 117 guest rooms was constructed. The operational environment included delivery and retrieval tasks that could be performed by service robots, as well as complaint-handling tasks that were exclusively assigned to human employees. The hotel service process was formulated as a Markov Decision Process (MDP), and a Proximal Policy Optimization (PPO) algorithm was employed to learn adaptive task allocation policies for service robots. In addition, a hybrid collaboration framework was developed in which reinforcement learning-based robots operated alongside rule-based human employees. While the PPO agents dynamically selected tasks according to the current system state, human employees followed predefined rule-based dispatching policies, enabling the evaluation of collaborative decision-making between heterogeneous service agents. Numerical experiments were conducted under static, dynamic, and dynamic collaboration scenarios, and the proposed model was compared with a CPLEX-based optimization benchmark. The results demonstrate that the proposed approach effectively adapts to real-time task arrivals and changing operational conditions. Furthermore, the hybrid collaboration structure improved overall operational performance by combining the adaptability of reinforcement learning with the stability and predictability of rule-based decision-making. The findings suggest that reinforcement learning can serve as an effective decision-making framework for dynamic task allocation in hotel service operations and that collaboration between learning-based robots and rule-based human workers can further enhance operational efficiency in complex service environments.

In the hotel industry, service quality is an important factor because it influences key outcomes such as guest satisfaction, reputation, and demand. High service quality can enhance a hotel’s reputation and lead to an increase in repeat customers and additional demand. Service quality depends on how hotels invest in their employees. Employee training can enhance employees’ service-related capabilities, while pay-for-performance can motivate employees to maintain high service quality. Based on this relationship, this study develops a Stackelberg game-theoretic model to examine hotels’ strategic decisions on employee training and pay-for-performance for long-term profit maximization. The model provides a framework for explaining how hotels may determine training and pay-for-performance levels over time. In the model, the hotel acts as the leader, setting two strategic decision variables: training investment and pay-for-performance. The employee acts as the follower and chooses an effort level based on individual utility. Reputation is modeled as a dynamic state variable. Higher service quality in one period accumulates as reputation, which in turn generates demand in subsequent periods. A central contribution of this study is that it jointly examines two key investments in hotels. Employee training improves employees’ service-related capabilities, and its effects can persist over time by contributing to reputation and future demand. Pay-for-performance can motivate employees to exert greater effort; however, as employees’ abilities and performance increase, hotels need to provide compensation that reflects their higher contribution. Therefore, this study proposes a decision-making model that jointly considers the interaction between these two factors. Overall, the proposed model serves as a hotel-based human-resource investment framework that considers both long-term capability development and short-term effort incentives, offering managerial implications for designing contracts that improve service quality, reputation, demand, and hotel profitability over time.

Demand-Responsive Transit (DRT) operates without fixed routes or timetables. Vehicles are dispatched in response to passengers' real-time requests. DRT is well suited to environments where demand is temporally concentrated, spatially dispersed, and highly variable. These characteristics align with the tourism context, where attractions are spread out and travel demand fluctuates considerably. DRT can serve as an alternative. It enables tourists to flexibly visit multiple dispersed attractions, even in transit-underserved regions with limited public transportation. However, realizing the potential of DRT requires sophisticated operational decisions. Operators must decide which vehicle serves which request, in what sequence, and when. This challenge intensifies because tourist travel requests arrive sequentially throughout operating periods. Operators must periodically revise plans while ensuring service continuity for passengers already onboard. This study develops an optimization model for DRT operations under such dynamic conditions. The problem is formulated as a mixed-integer linear programming (MILP) model. It is based on the standard structure of the vehicle routing problem (VRP) and the dial-a-ride problem (DARP) and is extended to account for mid-operation re-planning. The model fixes the assigned vehicle for onboard requests to guarantee service continuity. Vehicles resume operation from their current locations rather than the depot and may depart early. This static model is applied repeatedly via a rolling-horizon procedure. The system updates vehicle states and request sets at fixed intervals. It responds to dynamically arriving demand while preserving solution quality at each stage. The model is implemented using IBM ILOG CPLEX and validated through a small-scale numerical example. The results confirm that the re-optimization procedure maintains the continuity of onboard passengers while efficiently integrating newly arriving requests, with idle vehicles dispatched only when additional capacity is required. This study thus provides a foundation for real-time DRT operational decision-making that supports access to dispersed attractions in transit-underserved tourism settings.