Quadruped robots hit last‑mile

Quadruped robots hit last-mile A delivery robot that can climb stairs sounds like a gimmick until you remember where most deliveries still fail: not on the highway, but at the curb, the porch step, the apartment entry, or the muddy path between the van and the door. That short handoff is one of the most expensive and labor-intensive parts of the entire logistics chain, and it is exactly where quadruped robots are starting to show up in North America. The phrase “last mile” describes the final leg of delivery from a local hub to the customer. In practice, the hardest part is often even shorter than a mile: operators now talk about the “last meter” or “last 100 yards,” meaning the final stretch from street or van to doorstep. That is where sidewalks end, gates appear, curbs interrupt wheels, and stairs turn a smooth route into a manual job. Traditional sidewalk robots are built like coolers on wheels. They work best on flat paths, controlled campuses, and predictable pavement, which is why many early commercial deployments concentrated in university settings and dense neighborhoods with relatively uniform infrastructure. Once the route includes steep ramps, broken sidewalks, porch steps, or narrow passages, those robots can stall or require human help. Quadruped robots attack that problem by changing the shape of the machine. Instead of relying only on wheels, they use four articulated legs, or in some newer designs, a hybrid wheel-leg system that rolls when the ground is easy and walks when the environment gets awkward. That makes them less like a sidewalk cart and more like a pack animal built for curbs, stairs, and uneven ground. One of the clearest recent signals came from DEEP Robotics, which published a North America delivery video on April 7, 2026, saying its quadruped robots are already deployed there for autonomous delivery over complex terrain. The company has also described a logistics use case for its Lynx M20 Pro, a hybrid wheeled-legged robot aimed at the exact failure point where standard wheeled robots get stuck. In DEEP Robotics’ own specifications for that case, the Lynx M20 Pro can roll at up to 5 meters per second, climb obstacles up to 25 centimeters high, handle slopes up to 45 degrees, and carry payloads up to 15 kilograms. The company says the machine switches modes as terrain changes: wheels for speed on flat ground, legs for staircases and curbs. Those numbers matter because delivery routes are full of small vertical barriers. DEEP Robotics says many standard wheeled autonomous mobile robots struggle with obstacles above roughly 10 to 15 centimeters, which means a single curb or porch step can break the automation chain and force a person to finish the job. A robot that can clear 25 centimeters does not solve every route, but it covers many of the obstacles that make suburban and urban deliveries messy. Weather is another part of the pitch. DEEP Robotics says its logistics platform maintained operations in heavy rain with an Ingress Protection rating of IP66 and a loaded runtime of about 2.5 hours using hot-swappable batteries. That is a practical detail, not a flashy one, but rain and battery downtime are exactly the kind of operational friction that turns a pilot into a headache. This is not the only company trying to automate the awkward end of delivery. In May 2025, U.S. parcel platform Veho and Swiss robotics company RIVR launched a pilot in Austin, Texas, using a robot designed for the “last 100 yards” from vehicle to doorstep. Veho said the robot could navigate stairs, gates, and sidewalks while a human driver-partner and RIVR staff supervised the early deployment. That Austin pilot is useful context because it shows where the market is heading. Even when companies are not using a pure quadruped form, they are moving toward legged or wheel-leg machines because the old flat-ground model leaves too much of the real job undone. The commercial target is no longer just “a robot that can move on a sidewalk.” It is “a robot that can complete the handoff without calling a human for the last few steps.” The cost logic is straightforward. Vans and line-haul networks already move parcels efficiently over long distances, but the final handoff is slow, repetitive, and full of edge cases. If a robot can take over only the difficult tail end of selected routes, operators may not need to automate every delivery to save money; they just need to reduce the number of human minutes spent on the least efficient stops. That is an inference from the deployment claims and pilot structure, not a published industry-wide savings figure. The access argument may be just as important as the cost argument. A robot that can handle stairs, ramps, rough pavement, or wet ground opens up deliveries in places where small wheeled bots are unreliable: hillside housing, mixed-surface industrial sites, temporary event spaces, and residential areas with inconsistent infrastructure. That does not mean quadrupeds replace vans or people; it means they fill a gap that other delivery machines were never very good at crossing. There are still obvious limits. Battery life, payload size, safety oversight, local regulation, and customer acceptance all matter, and many deployments remain supervised or narrowly scoped rather than fully autonomous at city scale. Thomasnet’s 2025 overview of the sector noted that delivery robots have moved beyond pure testing in some environments, but the strongest results are still concentrated in short-distance, controlled, or highly specific use cases. What changed in this story is not that robots can carry packages. That has been true for years. What changed is that companies are now pushing machines built for awkward terrain into real North American last-mile work, where stairs, curbs, gates, and weather have always been the reason a person had to step in. If those deployments hold up outside demo videos and supervised pilots, the “delivery robot” category may start to split in two: simple wheeled bots for smooth ground, and quadruped or wheel-leg robots for the places where the route stops being a path and starts being an obstacle course. (shipveho.com/blog/veho-x