Research

Research

Modeling Stochastic Service Time for Complex On-Demand Food Delivery

The on-demand food delivery (OFD) industry faces significant challenges in accurately predicting service times, which are influenced by various uncertain factors. This paper proposes a Gaussian mixture model (GMM) and a hybrid estimation of distribution algorithm (HEDA) to estimate stochastic service times, improving decision-making and efficiency. The authors validate their approach through extensive offline and online experiments, demonstrating its effectiveness in real-world applications.

2026-08-13

Research

A 25% Reduction in Order Cancellations via Multi-Stage Bonus Allocation

This paper introduces a Multi-Stage Bonus Allocation (MSBA) framework for meal delivery platforms, aiming to reduce order cancellations and improve driver acceptance rates. The framework, consisting of an acceptance probability model, a Lagrangian dual-based dynamic programming (LDDP) algorithm, and an online allocation algorithm, was tested on the Meituan platform, resulting in a 25% reduction in canceled orders and a 30% savings in compensation for food waste.

2026-08-13

Research Digital, Intelligence & Platforms

Enhancing RL Generalization with Compositional Causal Components

The paper "Modeling Unseen Environments with Language-guided Composable Causal Components in Reinforcement Learning" by Xinyue Wang and Biwei Huang introduces a novel framework, WM3C, that enhances reinforcement learning (RL) generalization by decomposing tasks into composable causal components. This approach leverages language to guide the decomposition of the latent space, leading to better generalization in unseen environments.

2026-08-12

Research Digital, Intelligence & Platforms

Algorithmic Labor and Temporality in Chinese Food Delivery: A Study of Platform Workers

Ping Sun's study explores the labor conditions of delivery workers on food delivery platforms in China, focusing on how they experience and make sense of algorithms. The research reveals that these workers are not passive entities but active participants in creating their own "organic algorithms" to manage and sometimes subvert the system.

2026-08-12

Research Fulfillment & Last-Mile

Data-Driven Optimization Enhances Last-Mile Delivery by 5%

This paper, authored by Hongrui Chu, Wensi Zhang, Pengfei Bai, and Yahong Chen, introduces a data-driven optimization approach that combines machine learning with capacitated vehicle routing to improve the efficiency and timeliness of last-mile delivery (LMD). The proposed smart predict-then-optimize (SPO) framework outperforms traditional methods, reducing total delivery time and operating costs.

2026-08-12

Research Demand Forecasting & Inventory Management

N-BEATS: 11% Improvement in Time Series Forecasting

The N-BEATS architecture, proposed by Boris N. Oreshkin, Dmitri Carpov, Nicolas Chapados, and Yoshua Bengio, demonstrates that pure deep learning can outperform classical statistical methods in time series forecasting. The model, which is both interpretable and generic, achieved a 11% improvement over the statistical benchmark on M4 datasets.

2026-08-12

Research Digital, Intelligence & Platforms

Representation Learning Advances: 30% Improvement in Speech Recognition

The paper by Yoshua Bengio, Aaron Courville, and Pascal Vincent reviews recent advancements in representation learning and deep learning. It highlights significant improvements in speech recognition, natural language processing, and object recognition, demonstrating the potential of these techniques to revolutionize AI and supply chain applications.

2026-08-12

Research Fulfillment & Last-Mile

Optimal Infrastructure Planning and Order Assignment for Drone-Assisted Food Delivery

This paper investigates the optimal infrastructure planning and order assignment for on-demand food-delivery platforms using a mixed fleet of drones and human couriers. The study reveals that integrating drone delivery can reduce operational costs, decrease courier fleet size, and increase order bundling opportunities, but it also highlights the trade-offs between travel time savings and detours.

2026-08-12

Research Digital, Intelligence & Platforms

Meituan Cuts Order Cancellations by 25% with AI Bonus Framework

Meituan’s AI-powered Multi-Stage Bonus Allocation framework cut daily order cancellations by over 25%, reducing 165,000 NA-canceled orders and cutting restaurant food waste compensation by 30%. Built with a semi-black-box model and Lagrangian dual-based dynamic programming, it makes millisecond bonus decisions across order lifecycle stages. Deployed at scale, it demonstrates how real-time incentive optimization strengthens last-mile supply chain resilience — a lesson increasingly relevant for e-commerce logistics, perishable goods distribution, and time-sensitive service platforms globally.

2026-03-30

Research Digital, Intelligence & Platforms

**A Framework for Multi-Stage Bonus Allocation in Meal Delivery Platforms: Operationalizing Real-Time Incentive Optimization at Scale**

*SCI.AI Research Section — In-Depth Technical & Business Analysis* --- ### **Research Background: The Hidden Crisis of Order Cancellation in On-Demand Food Delivery** Food delivery platforms operat...

2026-03-26

Research Fulfillment & Last-Mile

When Uncertainty Becomes Productivity: Tsinghua-Meituan Research Reveals the Probabilistic Modeling Revolution in On-Demand Food Delivery Service Time

Tsinghua University and Meituan collaborative research proposes Gaussian Mixture Model and Hybrid Estimation of Distribution Algorithm, upgrading on-demand food delivery service time from deterministic assumptions to probabilistic distribution modeling. Rigorous online A/B tests verify the quantifiable business value of uncertainty modeling in improving ETA accuracy, optimizing dispatch decisions, and enhancing user experience.

2026-03-22

Research

Battery-Swapping Heavy Trucks in Thailand: A Supply Chain Inflection Point for ASEAN Electrification

U POWER’s deployment of 30 battery-swapping electric heavy trucks in Thailand—kicking off a 1,000-vehicle plan—is a pivotal inflection point for ASEAN supply chains. Unlike plug-in EVs, this model bypasses grid constraints, slashes downtime to under 11 minutes per swap, and enables closed-loop battery recycling with 92% mandated recovery by 2028. Thailand’s dual-track policy—excluding heavy trucks from subsidies while funding swap infrastructure—has catalyzed interoperable standards, ASEAN-wide regulatory alignment, and $1.2 billion in regionally anchored investment. The initiative signals a shift from import-dependent electrification to sovereign, digitally integrated freight infrastructure.

2026-03-20