Trips can be made more efficient and therefore more sustainable by tying them together with an intelligent agent. Bullit developed a proof-of-concept for Emons Group, a niche player in the European transport market that focuses on innovation and sustainable transport solutions.
Dataset
A dataset containing a list of trips was used, on which carriers can bid. The dataset contains about a thousand routes, including the pickup location, delivery location, frequency per year and distance in kilometres, among other things.
The choice was made to cluster the locations since the possibilities for the agent increase exponentially with each additional location. Combining this with the data on how many trips are leaving or arriving provides an initial indication of the pickup and delivery clusters where trucks are likely empty.
Reinforcement learning
The proof-of-concept used reinforcement learning, a technique within machine learning that allows an agent to learn from simulations. When training, the agent starts at a random pickup location each time and undertakes actions for which it receives scores. The agent keeps adjusting its neutral network based on those scores so that it makes better predictions in the future. On their website, Bullit Digital explains this approach step by step.
A negative score can be turned into a positive one after just ten minutes of training. The system is still learning new things after four thousand iterations, as the average score keeps rising slowly.
Want to know more about what AI can do for logistics? Contact us here.
