Many processes in logiscs are managed manually and solved empirically or through the use of heuristics. In the area of AI, many of the processes could be automized/improved through AI algorithms. In particualr we are interested in exploring the use of Reinforcement Learning (RL) algorithms to fullfil such purpose.
Using RL is interesting for two different reasosns; On the one hand, the many RL algorithms (PPO, TRPO, A3C) open a wide range the possibilities and versatility as candidates to find a solution for logistics problems. On the other hand, the different extensions of RL (and in particular multi-objective and multi-agent) can be combined to propose holistic interaction and learned behavior of complex tasks.
The project will be developed to solve the problem of pick-up and delivery optimization within a warehouse under different demand parameters, packaging restrictions, and external delivery. The objective of the developed agent will be to maximize the throughput of warehouse, while minimizing the resources used (time, distance covered by workers, redundancy of trips).
The validation plan of the thesis will take place using real-world datasets.
The implementation of this work will be based on the development of DQN-based algorithms to find a solutiion for the problem
n.cardozo