Humanoid robots to join Amazon’s last-mile race?

What is last-mile delivery?

In logistics, the ‘last mile’ means the final leg of a product’s journey – from a warehouse or local depot to the customer’s doorstep. It is considered to be one the most expensive and operationally challenging parts of the supply chain. When you’re running an operation that must account for variables like traffic, weather, customer availability, and delivery density, it can make optimisation difficult.

Over the last few decades, next-day and even same-day delivery expectations have grown significantly, and companies like Amazon who offer this service are under pressure to innovate. So, Amazon’s reported, though unconfirmed, test of robots capable of physically delivering parcels to doorsteps –  as in, stepping out of a van and up to a front door – could relieve pressure on human drivers by reducing the need for them to repeatedly exit their vehicles.

Amazon’s humanoid test facility

According to ‘The Information’, Amazon has begun work on a test site in its San Francisco offices, and it is described as a "humanoid park" designed to simulate the real-world challenges faced by delivery agents. These include stairs, doorsteps, and uneven terrain – all the sorts of variables a delivery robot might encounter at the customer end of a transaction.

The robots themselves will not be developed in-house. Amazon is reportedly sourcing them from third-party robotics companies such as Agility Robotics, whose Digit robot features legs and arms, and Unitree, a Chinese firm noted for its cost-effective humanoid designs. Instead, Amazon’s internal team will focus on developing the artificial intelligence and control systems required to help the robots operate autonomously and safely in unpredictable settings.

Amazon reportedly plans to send these robots out as support to existing delivery drivers, testing how well the machines can integrate with Rivian electric vans and hand parcels over at the final point of contact.

Navigating complex terrain

Over the past decade, Amazon has introduced numerous automation technologies within its fulfilment centres – including mobile robots for storage and retrieval, robotic arms for sorting and picking, and vision-guided systems for quality control.

The move into unstructured environments, like pavements and gardens, introduces new engineering and safety challenges because unlike warehouse floors, domestic spaces are unpredictable. Pets, children, steps, mud, loose gravel, and tightly packed corridors are all barriers to navigation and safe operation.

However, rather than leaping straight into general-purpose robotics, Amazon appears to be following a structured and scalable path toward automation. The speculated rollout of its humanoid delivery robots can be seen as the inevitable next step in a deliberately phased journey.

If we take a look back, that journey perhaps began with warehouse automation, notably following Amazon’s 2012 acquisition of Kiva Systems. Mobile robots were deployed to shuttle goods around fulfilment centres, and it set the standard of safe, repetitive movement in controlled environments.

This was followed by the company’s investment in simulation and AI training, including AWS DeepRacer – a reinforcement learning platform that gave developers a playground to train and refine autonomous driving agents. DeepRacer showed how software agents could be trained in virtual environments before being deployed in the real world – possibly not to dissimilar to that of the tests now reportedly taking place at humanoid park.

Meanwhile, Amazon has launched internal reskilling programmes to prepare staff for its automation future. Its Mechatronics and Robotics Apprenticeship pathway, for example, helps warehouse workers retrain as robotics technicians, contributing to a growing in-house capability around robotics engineering.

A deliberate path to embodied AI

From simulation tools and warehouse arms to humanoid delivery robots and reskilling technicians, Amazon’s seems to be taking a slow, deliberate, and methodical approach to efficiently scaling its operations.

Each new phase seems to be testing a slightly more complex environment – from warehouse floors to outdoor walkways, and now to customers' homes. As Amazon moves into this next stage of field testing, the focus will likely remain on practical deployment, cost-effectiveness, and how well human workers can collaborate with their robotic counterparts.

The last mile may be one of the hardest to conquer – but for Amazon, it is also a proving ground for the future of autonomous systems.