Interactive Fleet Learning

In the last few years, robotics and artificial intelligence have seen an exciting development with large fleets of robots leaving the lab and entering the real world. These robots use deep learning to operate autonomously in unstructured environments and can offload data, memory, and computation to the cloud via the Internet, a process known as “Fleet Learning”. However, these data-driven approaches face the problem of the “long tail”, where robots encounter new scenarios and edge cases not represented in the dataset. To ensure sufficient reliability for their services, companies can fall back on remote humans over the Internet who can interactively take control and “tele-operate” the system.