Foresee and Act Ahead: Task Prediction and Pre-Scheduling Enabled Efficient Robotic Warehousing

TL;DR

This work tackles inefficiencies in multi-robot warehousing caused by relying solely on current demand. It introduces a pre-scheduling framework that predicts task flow and allocates hybrid tasks, powered by the Temporal Decoupled Tri-Spatial Graph Convolutional Network (TDTGCN) and the Multi-Robot Hybrid Task Allocation (MR-HTA) with Hybrid-KM, including a domain-transfer embedding and a 3D spatio-temporal graph backbone. The authors demonstrate SOTA prediction accuracy on three real-world datasets and show lifelong validation in a factory with hundreds of robots, achieving over 50% improvements in metrics such as the empty running rate and mean pickup time, while maintaining near-optimal path efficiency. The approach enables proactive, balanced robot utilization and scalable pre-scheduling in warehousing, with strong implications for throughput and energy efficiency.

Abstract

In warehousing systems, to enhance logistical efficiency amid surging demand volumes, much focus is placed on how to reasonably allocate tasks to robots. However, the robots labor is still inevitably wasted to some extent. In response to this, we propose a pre-scheduling enhanced warehousing framework that predicts task flow and acts in advance. It consists of task flow prediction and hybrid tasks allocation. For task prediction, we notice that it is possible to provide a spatio-temporal representation of task flow, so we introduce a periodicity-decoupled mechanism tailored for the generation patterns of aggregated orders, and then further extract spatial features of task distribution with novel combination of graph structures. In hybrid tasks allocation, we consider the known tasks and predicted future tasks simultaneously and optimize the allocation dynamically. In addition, we consider factors such as predicted task uncertainty and sector-level efficiency evaluation in warehousing to realize more balanced and rational allocations. We validate our task prediction model across actual datasets derived from real factories, achieving SOTA performance. Furthermore, we implement our compelte scheduling system in a real-world robotic warehouse for months of lifelong validation, demonstrating large improvements in key metrics of warehousing, such as empty running rate, by more than 50%.

Figures (5)

• Figure 1: The pre-scheduling enhanced framework for robotic warehousing. It predicts future tasks along with the timeline while assessing the uncertainties, following which, a hybrid allocator is designed to allocate hybrid tasks. This allows otherwise idle robots to move ahead and be dynamically re-assigned them in each round of allocation, further exploring the potential of robotic labor in multi-robot warehousing system.
• Figure 2: The execution process of our complete scheduling framework.
• Figure 3: The predicted tasks are distributed such that they appear in each sector of $G_H$, rather than being specific to each node within $G$(left). And their spatio-temporal distribution dependencies are extracted through iterative extraction and integration using three types of graph structures(right).
• Figure 4: The architecture of the TDTGCN (shown on the right), main blocks connected by residuals. It processes historical task flow data $X_{h} \in \mathbb{R}^{R \times I \times F}$, initially through the MSTSF Block (as depicted on the left). This block embeds sparse data and decouples time-series patterns using multi-scale DWT and FFT. Subsequently, the data is processed by 3D Spatio-Temporal GCN Blocks, extracting spatio-temporal dependencies across 3 spatial dimensions. Finally, the features are integrated and a reverse mapping operation is performed to generate the predicted task flow $X_{p} \in \mathbb{R}^{R \times T' \times F}$.
• Figure 5: The layout of real-world factory for validation experiments