Driven by shifts in consumer behavior and the pandemic, e-commerce continues its explosive growth and transformation. As a result, logistics and transportation firms find themselves at the forefront of a parcel delivery revolution. This new reality is especially evident in last-mile delivery, which is now the most expensive element of supply chain logistics. It represents more than 41% of total supply chain costs across industries, from retail to manufacturing. Understandably, the surging costs of last-mile delivery are prompting efforts to identify and mitigate the underlying causes.
The last-mile delivery challenge is further complicated by the vehicle routing problem (VRP). A generalization of the traveling salesman problem, VRP asks, “What is the optimal set of routes that a fleet of vehicles should undertake to make deliveries to a specific set of customers?” With just 10 delivery destinations, over 3 million permutations and combinations of trips are possible. With 15 destinations, the number of possible routes can exceed 1 trillion. As the number of destinations increases, the corresponding number of possible trips surpasses even the capabilities of the fastest supercomputers. And this doesn’t account for common operational constraints, like fleet availability, navigation capabilities, and access limitations.
Transforming routing services
These constraints, along with possible adaptations and the constant evolution within the transportation and logistics field, make it increasingly challenging for businesses to establish, or even outsource, effective route optimization services.
clicOH, a member of the NVIDIA Inception program for startups, has developed a proprietary routing model to address these challenges. Its solution leverages the latest technologies from NVIDIA, from heuristic and metaheuristic optimization algorithms to machine learning and AI. And, by relying on the efficiency of NVIDIA libraries, clicOH’s application quickly adapts to the different requirements in package distribution density, cost efficiency, and delivery time optimization for last-mile delivery.
Optimizing last-mile delivery costs
clicOH aims to address a range of routing challenges using NVIDIA libraries. For example, the company adopted NVIDIA cuOpt to support its work related to the traveling salesman problem and to determine optimal delivery routes. The cuOpt library works with GPUs and libraries like RAPIDS and CUDA to generate faster and more accurate delivery routes.
Additionally, RAPIDS enables clicOH to implement unsupervised machine learning algorithms without the need to modify code, resulting in more efficient data analyses. These unsupervised algorithms enable the clustering of high-demand zip codes for more efficient delivery, as well as the identification of hard-to-reach areas. When combined with NVIDIA cuOpt, these algorithms can process thousands of routings in minutes or even seconds, optimizing delivery times while accounting for local routing constraints. This ultimately reduces delivery costs.
Using NVIDIA GPUs on AWS development environments, clicOH analyzed thousands of pre-existing routes across multiple cities to map routing inefficiencies. This analysis enabled clicOH to streamline the development of its logistics solution and enhanced the application’s adaptive capabilities.
clicOH has also developed a deep learning model to optimize delivery times, maximize fleet utilization, and identify zip codes that experience delivery challenges due to scheduling constraints. By optimizing its AI models with NVIDIA accelerated computing, clicOH has achieved a 20x speedup in cluster route planning and a 15% reduction in overall operating costs.
Learn more about clicOH accelerated logistics solutions. To further explore how NVIDIA cuOpt can enhance your fleet routing workflows, visit the NVIDIA Developer forums.
