Tomás Lagos
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Tomás Lagos has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Oleg A. Prokopyev
Department of Business Administration, University of Zurich, 8032 Zurich, Switzerland
3 shared publications Brittany Segundo
Wm Michael Barnes '64 Department of Industrial and Systems Engineering, Texas A&M University, College Station, Texas, United States of America
2 shared publications- Felipe Lagos
Faculty of Engineering and Sciences, Universidad Adolfo Ibáñez, Santiago de Chile 7941169, Chile
2 shared publications Junyeong Choi
Department of Ecology and Conservation Biology, Texas A&M University, College Station, Texas, United States of America
2 shared publicationsLewis Ntaimo
Wm Michael Barnes '64 Department of Industrial and Systems Engineering, Texas A&M University, College Station, Texas, United States of America
2 shared publications- Ramon Auad
Amazon.com, Bellevue, Washington 98004; and Universidad Católica del Norte, Antofagasta 1240000, Chile
2 shared publications
Latest Publications
- 27 April 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
Data-Driven Optimization for Meal Delivery: A Reinforcement Learning Approach for Order-Courier Assignment and Routing at Meituan
The rapid growth of online meal delivery has introduced complex logistical challenges, where platforms must dynamically assign orders to couriers while accounting for demand uncertainty, courier autonomy, and service efficiency. Traditional dispatching methods, often focused on short-term cost minimization, fail to capture the long-term implications of assignment decisions on system-wide performance. This paper presents a novel hybrid framework that integrates reinforcement learning with hyper-heuristic optimization to improve sequential order assignment and routing decisions in meal delivery operations. Our approach combines n-step state-action-reward-state-action with value function approximation and a multiarmed bandit-based hyper-heuristic incorporating seven specialized low-level heuristics. Our approach explicitly models the evolving system state, enabling dispatching policies that balance immediate efficiency with future operational performance. By employing scalable linear value function approximation, we enhance policy learning in high-dimensional environments while maintaining generalization across states and actions. Using real operational data from the food delivery platform Meituan, we develop a comprehensive simulation environment that captures order dynamics, courier behavior, and service times. Through extensive computational experiments, we demonstrate that our framework significantly outperforms traditional benchmark policies, achieving 12% cost reduction through strategic order postponement. Our results reveal that the largest improvements occur during high-demand periods with courier shortages and that a 10% increase in courier availability yields greater benefits than algorithmic improvements alone. The proposed methodology effectively balances immediate operational efficiency with long-term performance while providing valuable insights for meal delivery platforms regarding courier fleet management and order assignment strategies. History: This paper has been accepted for the Transportation Science Special Issue The First INFORMS TSL Data-Driven Research Challenge.
- 27 April 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
Data-Driven Optimization for Meal Delivery: A Reinforcement Learning Approach for Order-Courier Assignment and Routing at Meituan
The rapid growth of online meal delivery has introduced complex logistical challenges, where platforms must dynamically assign orders to couriers while accounting for demand uncertainty, courier autonomy, and service efficiency. Traditional dispatching methods, often focused on short-term cost minimization, fail to capture the long-term implications of assignment decisions on system-wide performance. This paper presents a novel hybrid framework that integrates reinforcement learning with hyper-heuristic optimization to improve sequential order assignment and routing decisions in meal delivery operations. Our approach combines n-step state-action-reward-state-action with value function approximation and a multiarmed bandit-based hyper-heuristic incorporating seven specialized low-level heuristics. Our approach explicitly models the evolving system state, enabling dispatching policies that balance immediate efficiency with future operational performance. By employing scalable linear value function approximation, we enhance policy learning in high-dimensional environments while maintaining generalization across states and actions. Using real operational data from the food delivery platform Meituan, we develop a comprehensive simulation environment that captures order dynamics, courier behavior, and service times. Through extensive computational experiments, we demonstrate that our framework significantly outperforms traditional benchmark policies, achieving 12% cost reduction through strategic order postponement. Our results reveal that the largest improvements occur during high-demand periods with courier shortages and that a 10% increase in courier availability yields greater benefits than algorithmic improvements alone. The proposed methodology effectively balances immediate operational efficiency with long-term performance while providing valuable insights for meal delivery platforms regarding courier fleet management and order assignment strategies. History: This paper has been accepted for the Transportation Science Special Issue The First INFORMS TSL Data-Driven Research Challenge.
- Open Access
- 13 February 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (2) , e0342049
Landscape-level effectiveness of fuel treatments in a forest-dominated ecosystem in the Southern United States
Wildfire has become an increasing threat to natural ecosystems and human livelihood alike in many parts of the world. Vegetation fuel treatments are considered a viable option for mitigating wildfire risk and damage; yet existing studies have yielded mixed or inconclusive results on fuel treatment effectiveness especially at the landscape level. Using fire behavior simulations and statistical analysis of simulation outputs, we assessed landscape-level effectiveness of prescribed burning (PB) and thinning from below (TFB) relative to their site-level effectiveness in terms of area burned (AB) and total cost of treatment and timber loss (TC) in a forest-dominated ecosystem in the southern United States. We found that effectiveness of a treatment varied with measurement metrics and extent, vegetation characteristics and dynamics, and their interactions with the treatment. PB and TFB were less effective at the landscape level than at the site level where fires burned only inside the treatment area. At both site and landscape levels, the effectiveness of PB and TFB in reducing AB and TC largely depended on the quantity of biomass and fire ignition location. TFB outperformed PB in mitigating both AB and TC with a larger timber volume, a longer delay in fire occurrence after treatment, or a higher uncertainty of fire ignition location. TFB was also more effective than PB in reducing TC at the landscape level. By clarifying the conditions under which a fuel treatment can mitigate the area burned and the total cost, this study advanced knowledge of fuel treatment effectiveness especially at the landscape level. Such knowledge can aid in developing and deploying treatment strategies to minimize fire extent and adverse economic consequences in the study region and beyond.
- 13 February 2026
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 21 (2) , e0342049
Open AccessLandscape-level effectiveness of fuel treatments in a forest-dominated ecosystem in the Southern United States
Wildfire has become an increasing threat to natural ecosystems and human livelihood alike in many parts of the world. Vegetation fuel treatments are considered a viable option for mitigating wildfire risk and damage; yet existing studies have yielded mixed or inconclusive results on fuel treatment effectiveness especially at the landscape level. Using fire behavior simulations and statistical analysis of simulation outputs, we assessed landscape-level effectiveness of prescribed burning (PB) and thinning from below (TFB) relative to their site-level effectiveness in terms of area burned (AB) and total cost of treatment and timber loss (TC) in a forest-dominated ecosystem in the southern United States. We found that effectiveness of a treatment varied with measurement metrics and extent, vegetation characteristics and dynamics, and their interactions with the treatment. PB and TFB were less effective at the landscape level than at the site level where fires burned only inside the treatment area. At both site and landscape levels, the effectiveness of PB and TFB in reducing AB and TC largely depended on the quantity of biomass and fire ignition location. TFB outperformed PB in mitigating both AB and TC with a larger timber volume, a longer delay in fire occurrence after treatment, or a higher uncertainty of fire ignition location. TFB was also more effective than PB in reducing TC at the landscape level. By clarifying the conditions under which a fuel treatment can mitigate the area burned and the total cost, this study advanced knowledge of fuel treatment effectiveness especially at the landscape level. Such knowledge can aid in developing and deploying treatment strategies to minimize fire extent and adverse economic consequences in the study region and beyond.
- 16 August 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
- Vol. 59 (1)
The Online Shortest Path Problem: Learning Travel Times Using a Multiarmed Bandit Framework
In the age of e-commerce, logistics companies often operate within extensive road networks without accurate knowledge of travel times for their specific fleet of vehicles. Moreover, millions of dollars are spent on routing services that fail to accurately capture the unique characteristics of the drivers and vehicles of the companies. In this work, we address the challenge faced by a logistic operator with limited travel time information, aiming to find the optimal expected shortest path between origin-destination pairs. We model this problem as an online shortest path problem, common to many last-mile routing settings; given a graph whose arcs’ travel times are stochastic and follow an unknown distribution, the objective is to find a vehicle route of minimum travel time from an origin to a destination. The planner progressively collects travel condition data as drivers complete their routes. Inspired by the combinatorial multiarmed bandit and kriging literature, we propose three methods with distinct features to effectively learn the optimal shortest path, highlighting the practical advantages of incorporating spatial correlation in the learning process. Our approach balances exploration (improving estimates for unexplored arcs) and exploitation (executing the minimum expected time path) using the Thompson sampling algorithm. In each iteration, our algorithm executes the path that minimizes the expected travel time based on data from a posterior distribution of the speeds of the arcs. We conduct a computational study comprising two settings: a set of four artificial instances and a real-life case study. The case study uses empirical data of taxis in the 17-km-radius area of the center of Beijing, encompassing Beijing’s “5th Ring Road.” In both settings, our algorithms demonstrate efficient and effective balancing of the exploration-exploitation trade-off.
- 16 August 2024
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Transportation Science
- Vol. 59 (1)
The Online Shortest Path Problem: Learning Travel Times Using a Multiarmed Bandit Framework
In the age of e-commerce, logistics companies often operate within extensive road networks without accurate knowledge of travel times for their specific fleet of vehicles. Moreover, millions of dollars are spent on routing services that fail to accurately capture the unique characteristics of the drivers and vehicles of the companies. In this work, we address the challenge faced by a logistic operator with limited travel time information, aiming to find the optimal expected shortest path between origin-destination pairs. We model this problem as an online shortest path problem, common to many last-mile routing settings; given a graph whose arcs’ travel times are stochastic and follow an unknown distribution, the objective is to find a vehicle route of minimum travel time from an origin to a destination. The planner progressively collects travel condition data as drivers complete their routes. Inspired by the combinatorial multiarmed bandit and kriging literature, we propose three methods with distinct features to effectively learn the optimal shortest path, highlighting the practical advantages of incorporating spatial correlation in the learning process. Our approach balances exploration (improving estimates for unexplored arcs) and exploitation (executing the minimum expected time path) using the Thompson sampling algorithm. In each iteration, our algorithm executes the path that minimizes the expected travel time based on data from a posterior distribution of the speeds of the arcs. We conduct a computational study comprising two settings: a set of four artificial instances and a real-life case study. The case study uses empirical data of taxis in the 17-km-radius area of the center of Beijing, encompassing Beijing’s “5th Ring Road.” In both settings, our algorithms demonstrate efficient and effective balancing of the exploration-exploitation trade-off.
- Open Access
- 14 July 2024
- Published by Elsevier in European Journal of Operational Research
- Vol. 320 (1) , 205-218
Bilevel optimization approach for fuel treatment planning
- 14 July 2024
- Published by Elsevier in European Journal of Operational Research
- Vol. 320 (1) , 205-218
Open AccessBilevel optimization approach for fuel treatment planning
- 4 October 2023
- Published by Elsevier in Operations Research Letters
- Vol. 51 (6) , 612-617
On complexity of finding strong-weak solutions in bilevel linear programming
- 4 October 2023
- Published by Elsevier in Operations Research Letters
- Vol. 51 (6) , 612-617
On complexity of finding strong-weak solutions in bilevel linear programming
- 14 February 2023
- Published by Springer Nature in Journal of Global Optimization
- Vol. 88 (1) , 199-232
Finding groups with maximum betweenness centrality via integer programming with random path sampling
- 14 February 2023
- Published by Springer Nature in Journal of Global Optimization
- Vol. 88 (1) , 199-232
Finding groups with maximum betweenness centrality via integer programming with random path sampling
- 1 August 2021
- Published by Elsevier in Resources Policy
- Vol. 72
Adaptive open-pit mining planning under geological uncertainty
- 1 August 2021
- Published by Elsevier in Resources Policy
- Vol. 72
Adaptive open-pit mining planning under geological uncertainty