Legged delivery robots: what actually delivers in 2026

In May 2019 Ford and Agility Robotics showed a concept that has shaped every presentation on this topic since: a self-driving vehicle stops at the kerb, a two-legged robot steps out of the boot and carries the parcel to the front door. In January 2020 Ford became the first customer and received the first two units. Then it went quiet. There has been no public update since, and Agility Robotics today lists warehouse and manufacturing work only on its own solutions pages. The vendor left the delivery market without ever announcing that it had.

Almost simultaneously, at CES 2019, Continental showed a shuttle releasing four-legged robots for the final metres. The image travelled the world. It never became a product, and ANYbotics, whose robot appeared in it, today sells industrial inspection for energy, mining and chemicals.

The pattern repeats, and it is worth knowing before you build a project on it: the demonstration is not the hard part.

In July 2026 Boston Dynamics published a post titled „Bridging the Porch Gap“. It shows a Spot hopping out of the back of a delivery vehicle and covering the last roughly fifty feet to the door, with a conveyor payload on its back and room for two parcels.

What matters is less the video than the sentence next to it. Boston Dynamics writes that today this is a demo. The company is explicitly seeking logistics partners and describes its goal as a full pilot with an industry leader, initially with human remote monitoring. No partner is named, and no timeline either.

That candour is a good sign from the manufacturer and an important signal for operators. If you receive an offer today that goes beyond this state of play, ask what it rests on.

RIVR, which came out of the Robotic Systems Lab at ETH Zurich and was initially known as Swiss-Mile, was the only European company seriously pursuing legged delivery. The technology is a wheel-leg hybrid: a four-legged robot with driven wheels that climbs stairs and kerbs and rolls on flat ground.

There were two documented pilots. From late March 2025 a trial ran with the British parcel carrier Evri in Barnsley, explicitly for the last hundred yards and explicitly alongside the courier rather than replacing them. In late May 2025 the first US deployment followed with Veho in Austin, supported by an operator on site and a remote monitoring team.

On 19 March 2026 Amazon acquired RIVR. Terms were not disclosed. At Amazon too the language is testing, not routine operation. For the European market that means the only domestic vendor in this category is now owned by a US corporation, and the technology is no longer generally available for purchase.

While the legged systems drew attention, a less spectacular form factor built volume. Starship Technologies reports over 10 million deliveries with its six-wheeled pavement robots, across more than 300 towns and campuses, with German partners including REWE and Merck.

This is not a statement about wheels being superior to legs. It is a statement about maturity. Wheels are simpler, cheaper, more robust and more energy efficient, and they handle the majority of any route. Legs solve the exception: the step, the kerb, the stairwell. That exception is real, but it is a smaller share of the route than the videos suggest.

So if you are planning today, invert the question: not which form factor is the most impressive, but what share of my route needs legs at all.

All of the pilots named above have one thing in common: they ran with a human alongside. That is often presented as a transitional arrangement, but it points to the actual problem. A robot can cross a car park and take a step. What it cannot do without integration is ring an intercom, trigger a door release, reach the third floor, prove who took the parcel, and deal with rain.

That is not a robotics question, it is a building-services question. And the documentation there is remarkably thin: no delivery-robot manufacturer publishes a protocol for intercom and door-entry systems or names a door-system partner. It is the same pattern as the lift in hospitals and care homes, where no manufacturer publishes a lift protocol either.

The sober consequence for your business case: the door is a per-site integration item, not a product feature. If you cannot find it in the offer, the offer is incomplete.

It is the second reason things move more slowly here than the videos suggest. Under the German road traffic regulations a motor-driven robot is a vehicle, and section 2(1) StVO assigns vehicles to the carriageway. Running on the pavement therefore requires an individual exemption under section 46(2) StVO together with section 70(1) no. 2 StVZO, granted by a Land or municipal authority. There is no federal one-stop shop.

A common misconception is that the German autonomous driving act covers this. It does not: sections 1d to 1l StVG require the vehicle to be registered, and a robot with a design speed of up to 6 km/h falls outside the registration regime entirely under section 1 FZV. The condition cannot be met by construction.

What such a permit looks like in practice is visible in the Hamburg exemption for the Starship robots, public via a freedom-of-information request: one accompanying person per robot in public space, remote control, pavements at least one metre wide, at most fifteen robots per delivery area, operation suspended in fog and heavy rain, and an indemnity for the city. Plus the explicit statement that the permit sets no precedent for any later approval.

That this is not meant to remain the permanent state is now official: on 25 and 26 March 2026 the German Conference of Transport Ministers resolved to ask the federal government for a legal framework beyond the exemption regime. It has not been drafted yet.

Start on ground you own. A campus, a works site, hospital grounds, a business park or a new quarter give you a defined route, an answerable approval path and a genuine handover problem to solve. Everything you learn there transfers to the public realm later. The reverse does not hold.

Plan for remote supervision as a component rather than a defect. It is present in every documented pilot, and an offer that omits it is only deferring the cost.

And be sceptical of comparison tables quoting euros per delivery that feature a robot which does not yet exist. That can only be modelled with your volumes, your route and your labour cost.