Last Mile Route Optimization (also known as VRP) is a technique where you take a set of agents, say van drivers, and a set of customers, and you solve for the optimal fleet and routes to service them. Optimal routes are lower cost, lower distance, and shorter time than routes that a human planner could make for a big fleet. The classic example is called the traveling salesperson problem where you have a single agent that wants to find the fastest route between a set of customers. But modern businesses often have multiple agents and large numbers of customers, so we want to automate the planning of many optimal routes all at the same time. Keep in mind that we are also including many complex constraints such as the delivery time windows, vehicle capacity, driver shift time and breaks, volume, weight, and more. Now this works pretty well and is relatively fast with modern cloud software platforms up to 100 or even 1000 customers per day or shift, such as Onfleet, Workwave, OptimoRoute, Routific, etc.
However, route optimization for fleets with over 500 customers per day or shift takes on new challenges that traditional solutions struggle with. The problem is computing power. If there are only 4 stops, there are 24 unique routes that the person could travel to go through each of them once. If there are 10 stops, this number balloons to 3.6 million possible routes. For 5 vehicles that have to visit 50 stops, this is now an astounding 10 to the power of 227 which is more than the number of stars estimated in the universe. IBM’s SUMMIT super computer, the world’s fastest, would take 10 to the power of 203 years to get an exact solution to this problem.
So as you can imagine, the technologies based on these old algorithms don't scale well for large fleets. In order to have practical tools, these solutions take shortcuts and use assumptions and estimations.
A common example of these shortcuts is breaking the input datasets up by suburb or post code and optimizing them separately. This is a great way to reduce one big problem down into lots of smaller problems which are easier to solve, but you reduce the efficiency of the overall solution greatly. Another one is using the straight line distance between the points to process the optimization. But this is misleading because it doesn’t reflect a real road network. We all know that going 1 km between two points can take very different amounts of time depending on speed limits and road types, traffic, and more.
In addition to our advanced optimization engine, we have unique machine learning models that use real traffic data and cost KPIs to improve the algorithm over time. This is completely customized for each organization automatically to achieve the maximum benefits.
Using this API, you will be able to:
- Bundle orders from different customers into delivery routes at the lowest cost.
- Minimize the number of routes and therefore the number of required drivers.
- Determine the optimal order of visits in each route so that the duration is minimized while still meeting all time windows and other constraints.
- Support both pickup and delivery activities using the volume or weight capacity of the vehicle constantly.
- Assign the right inventory to depots, dark stores, MFCs, or PUDOs based on flexible constraints such as time or distance.
- Simulate and execute a range of business models quickly and cost efficiently.
- If some orders cannot be included, they will be returned in a list of unassigned jobs.
Now you know why route optimization is so important and some history behind it. Next, check out real case studies from our customers and explore our API.
