Coordinating hundreds of robots at once is a major challenge. Inside an automated warehouse, robots constantly receive new tasks and must be redirected quickly as they travel along aisles, collecting and delivering products to fulfill customer orders. Even minor traffic jams can quickly escalate, causing widespread delays. To avoid these slowdowns, MIT and Symbotic researchers developed a new system that learns which robots should go first, based on congestion, and reroutes them in advance to prevent bottlenecks.
Traditional systems use algorithms written by human experts to coordinate robot movement. However, congestion or collisions can force costly shutdowns for manual fixes. To achieve adaptability, the MIT team designed a neural network that observes the warehouse and prioritizes robots using deep reinforcement learning, combined with a fast planning algorithm that helps robots respond quickly to changing conditions. The model learns through trial and error in warehouse simulations, earning rewards for boosting throughput and avoiding conflicts. Over time, the model will be able to coordinate a fleet of robots.
“This is a very promising approach, because in these giant warehouses even a 2 or 3 percent increase in throughput can have a huge impact,” said Han Zheng, a graduate student in the Laboratory for Information and Decision Systems (LIDS) at MIT and lead author of a paper on this new approach. “The planning system needs to be adaptive to these changes as the warehouse operations go on. By interacting with simulations inspired by real warehouse layouts, our system receives feedback that we use to make its decision-making more intelligent. The trained neural network can then adapt to warehouses with different layouts.”
The system accounts for obstacles and dynamic interactions between robots as they move. Predicting robot interactions avoids congestion before it occurs. Once the neural network assigns priority, a planning algorithm tells robots how to move, allowing quick reactions to changing conditions. This hybrid approach is essential. “Pure machine-learning methods still struggle to solve complex optimization problems, and yet it is extremely time- and labor-intensive for human experts to design effective methods. But together, using expert-designed methods the right way can tremendously simplify the machine learning task,” said senior author Cathy Wu.
After training, the researchers tested the system in new simulated warehouses, designing their own environments to better mimic real operations. In simulations of real warehouses, this approach increased throughput by 25% compared to other methods and adapted well to different layouts and robot numbers. While not yet ready for real-world use, these results highlight the potential of machine learning in warehouse automation. The team plans to expand their system to larger warehouses and include task assignments for each robot.
