A single delivery-robot patent is a product idea; a family of them filed in one year is a strategy. In 2021 FedEx Corporate Services was granted a cluster covering a modular autonomous bot, including US11077551B2 (modular mobility base), US11090802B2 (multiple mobility base assembly), US11135717B2 (detachable autonomy-control module) and US11117255B2 (modular cargo cart).
The boring robot already shipped, and the interesting thing about FedEx's version is the seams. Across the family, the bot decomposes into swappable modules: a mobility base, a cargo container, an auxiliary power module, and a detachable autonomy-control 'brain.' Classified across B25J 9/163, B60W 30/09 (collision mitigation) and the G05D 1/00 autonomy series, the engineering is less about a clever sensor and more about a robot you can service like a truck.
ROI per square foot is the wrong frame for last-mile; the right one is cost per stop and uptime per unit. Modularity attacks uptime directly: if the autonomy brain fails, you swap the module rather than benching the whole robot, and a dead battery is a hot-swap, not a charging delay. For a logistics operator, that serviceability is the entire economic case.
The honest read is that a patent family is not a deployment. FedEx's Roxo bot generated headlines and then a quieter retreat, and the filings outlasted the most visible pilots. That is the recurring pattern in last-mile robotics: the IP describes a serviceable, modular vision while the unit economics of actual sidewalk delivery remain stubborn.
Still, the modularity thesis aged well even where FedEx's own program did not. Every serious delivery-robot operator since has converged on swappable batteries and field-replaceable compute, because a fleet that can only be repaired in a depot does not scale. FedEx patented the obvious-in-hindsight architecture early.
Backlog is the only honest demo, and the lesson from this family is to read logistics-robot patents for serviceability, not for the autonomy buzzwords. The company that can swap a module in the field beats the one with the cleverer perception stack but a robot that has to be trucked back for every fault.