Amazon has spent more than a decade automating its logistics network. The company has already deployed more than 1 million robots across its operations, and its recent acquisitions have steadily extended machine capability deeper into the fulfillment stack. What those investments did not solve was what happens after a package leaves the van.
RIVR, an ETH Zurich spinout, develops four-legged robots that can navigate real-world environments, including steps, curbs, and uneven terrain, and are designed to accompany human drivers through the final stage of delivery. Amazon confirmed the acquisition to its third-party delivery contractors in a notice obtained by CNBC, framing the move as a safety investment.
“We believe this technology, when working alongside your delivery associates, has the potential to further improve safety outcomes and the overall customer experience, particularly in the last steps of the delivery process,” Amazon wrote.
The commercial logic is harder than the safety language suggests. The last mile is the most expensive segment of any delivery route because it does not behave like a warehouse. Every stop involves different terrain, different access points, and different obstacles. That variability is precisely why human couriers remain the default even as the rest of the logistics chain is automated.
Amazon tried this before and failed. Its Scout delivery robot was discontinued after tests in several U.S. cities because a robot limited to flat sidewalks solves only a narrow version of the problem.
The sidewalk-only approach is also the model that Serve Robotics, spun out of Uber after its Postmates acquisition, has pursued at scale. Serve’s robots currently traverse several Los Angeles neighborhoods, rolling on sidewalks to pick up food orders from 900 restaurants within a two-mile radius, and the company is planning to deploy 2,000 Uber Eats delivery robots this year, expanding beyond Los Angeles to Dallas and other cities. Serve Robotics CEO Ali Kashani told PYMNTS that Uber CEO Dara Khosrowshahi has publicly said it costs less to use Serve’s robots than human couriers. The unit economics work on flat, predictable sidewalks in dense urban grids.
RIVR is built for everywhere else. The RIVR Two travels at up to 8.7 mph, carries more than 60 pounds of parcels, and can stop at red lights, open gates and climb stairs, as reported by SiliconANGLE.
Rather than operating independently on sidewalks, after reaching a delivery stop, the driver and robot can head off in different directions to handle drop-offs simultaneously, cutting time per stop without removing the human from the route. That hybrid model is a more conservative bet on near-term deployment than full autonomy, but it is also more compatible with the suburban and mixed-terrain environments where most Amazon volume lands.
Days after closing RIVR, Amazon announced a second robotics acquisition aimed at a different environment entirely, as reported by PYMNTS. Amazon acquired Fauna Robotics, a 2024-founded startup led by former Meta and Google engineers that builds humanoid robots designed to be approachable for consumers and businesses.
Too Many Robots
Inside the warehouse, a different constraint has come into focus. Researchers at MIT and the logistics technology firm Symbotic published that identifying fleet coordination, not robot capability, as what is now limiting throughput in large-scale automated fulfillment.
The underlying problem is straightforward. A warehouse running hundreds of robots simultaneously is essentially a traffic management challenge. The MIT system addresses this by having the software learn which robots should get priority at any given moment based on where congestion is forming, then rerouting others before the jam develops. The approach can quickly adapt to new environments with different numbers of robots or varied warehouse layouts, which matters because operators cannot rebuild their routing logic from scratch for every facility.
The system achieved roughly 25% higher throughput than traditional approaches in tests modeled after actual eCommerce warehouse layouts. As lead author Han Zheng noted, “in these giant warehouses, even a 2 or 3% increase in throughput can have a huge impact.”
