MIT researchers have designed an artificial intelligence solution to help robotic warehouses operate more efficiently. Automated warehouses, which employ hundreds of robots to pick and deliver goods, are becoming more commonplace, especially in industries such as e-commerce and automotive production. However, coordinating this robot workforce to avoid collisions, while also maintaining a high operational pace, presents a considerable challenge.
The researchers addressed this issue by applying techniques typically used to alleviate urban congestion to the problem. They created a deep-learning model that encodes complex details about the warehouse environment – including the robots, planned paths, tasks, and obstacles – and uses this information to identify the best warehouse locations to decongest, thus enhancing overall efficiency.
The innovative method groups the robots together and breaks down the complex task of coordinating them into smaller, manageable chunks. In practice, this approach allowed the researchers to manage congestion nearly four times faster than possible using a random search method.
Aside from improving warehouse operations, this approach could potentially be applied to other complex planning tasks, such as computer chip design or large-building pipe routing.
The novel neural network in use by the researchers is suitable for real-time operations due to its unique ability to efficiently encode extensive information about the robots in terms of their trajectories, origins, destinations, and relationships with other robots. This not only ensures that computations are reused across the groups of robots, but it also tackles key issues such as congestion management in a focused manner.
Upon testing their approach in various simulated environments, the researchers found that it managed congestion four times faster than traditional non-learning-based methods. Even when considering the extra computational burden of running the neural network, their approach still addressed the problem 3.5 times faster.
In the future, the team hopes to translate insights from their neural model into simple, rule-based methods to make implementation in actual robotic warehouses more straightforward.
The project, which was supported by Amazon and the MIT Amazon Science Hub, underscores the significant potential of AI in optimising operations across multiple industries. The results are particularly promising, not only in improving pre-existing methods in terms of solution quality and speed but also in displaying an impressive ability to generalize to previously unseen situations.
