AI system learns to keep warehouse robot traffic running smoothly

Source Domain: news.mit.edu

Researchers from MIT and Symbotic have developed a new method to maintain smooth operation in autonomous warehouses with hundreds of robots by preventing congestion and prioritizing mobile robots.
Their method uses deep reinforcement learning and a planning algorithm to predict and reroute robots to avoid bottlenecks, demonstrating about 25% greater throughput in simulations compared to traditional algorithms.
The hybrid system blends deep reinforcement learning for complex problem-solving with efficient planning algorithms to quickly adapt to changing warehouse conditions.
The system can effectively predict future robot interactions and adapt to various warehouse layouts, showing potential for significant performance gains even with minor improvements in throughput.
Despite not being yet deployed in real-world settings, the research demonstrates the promise of machine learning in optimizing warehouse operations and reducing inefficiencies.
The team plans to improve task assignments and scale the system for larger warehouses with more robots in future work.

AI system learns to keep warehouse robot traffic running smoothly | MIT News