The race toward last-mile automation is apparently back on.
Amazon’s new research lab just reported a big breakthrough that could accelerate the push for autonomous humanoid robots to your doorstep (and pretty much anywhere else). Meanwhile DoorDash is jumping into hardware with its new adorable mobile robot it hopes triggers new growth.
DoorDash’s Dot
Launched by Stanford students in 2013, DoorDash has become synonymous with on-demand food delivery. Its business model, heavily reliant on gig workers called Dashers, has reshaped how American perceive convenience in dining.
After years of impressive growth, the company faces fresh challenges, tighter margins, rising competition, and increasing regulatory pressure. Now publicly traded and headquartered in San Francisco, the company just officially unveiled its first self-developed delivery robot.
The autonomous wheeled machine, named Dot, features a rounded red body that makes it look approachable and non-threatening. Two LED lights resemble eyes, while a dome-shaped cap houses sensors and cameras. A LiDAR unit mounted on its top is for autonomous navigation. At about 137 cm (4.5 feet), Dot is roughly the height and footprint of a child’s stroller, though it’s much heavier at 159 kg (350 lbs).
DoorDash has been exploring last-mile automation since at least 2017, partnering with companies like GM’s self-driving Cruise unit and Google’s drone delivery arm, Wing. Dot, however, is its first in-house hardware product.
The pilot Dot program is running in Tempe and Mesa, AZ, with the potential to reach up to 1.5 million people by the end of the year.
According to DoorDash, Dot navigates complex urban environments and has a payload capacity of nearly 14 kg (30 lbs). That’s roughly the equivalent of six large pizzas. It reaches speeds up to 32 km/h (20 mph) and runs a reported 6 hours on a swappable battery.
The robot integrates into DoorDash’s new autonomous delivery platform, described as an AI-powered dispatcher that decides whether an order should be handled by a human Dasher or a machine.
The company says it’s completed hundreds of deliveries with the robot but declined to disclose how many units are operational. It describes the current phase as an early access program.
Despite internal optimism, challenges remain. Dot cannot climb stairs or use elevators, limiting its usefulness in apartment complexes and business parks. Questions also persist about municipal regulations, public safety, and risks of tampering and vandalism.
There are also concerns that robots like Dot could displace human delivery work, which many have grown to rely on during tough economic times. DoorDash insists its autonomous pursuits are not intended to replace human drivers, but to take on small delivery jobs that aren’t profitable for Dashers.
Whether that intention holds in practice is an open question.
Amazon FAR’s OmniRetargeting
Meanwhile, Amazon is taking big steps toward last-mile automation as it trains humanoids for real-world tasks.
The tech giant’s new Frontier AI Research lab, known as Amazon FAR, just unveiled a new system called OmniRetarget that makes it easier to teach robots new skills by watching humans. The researchers say the breakthrough tackles one of the biggest blockers to general-purpose robotics: the messy gap between human movement and robot execution.
According to the team’s paper, OmniRetarget focuses on the interactions that matter, like where the feet touch the ground, how the hands grip an object, and how the body maintains balance.
The lab shared footage of a Chinese Unitree G1 humanoid executing tasks from just a handful of human demonstrations. The robot performs full-body, interaction-heavy behaviors like carrying boxes across uneven terrain, climbing up platforms, and even parkour-style sequences with jumps and rolls.
With the OmniRetarget pipeline, the motions were cleaned up, retargeted, and then used to train reinforcement learning policies in simulation. Those policies transferred zero shot to the physical hardware.
As described, the framework basically supercharges the learning of skills that made Boston Dynamics’ Atlas humanoid robot go viral. It’s a step toward robots that can watch a human do something and then do it themselves pretty much right away.
Amazon launched the research lab earlier in 2025 as multiple news outlets reported it was developing software to control humanoids that could one day perform doorstep delivery. Engineers reportedly set up a humanoid park at a San Francisco facility for testing. The idea is that robots could ride in Amazons Rivian vans and spring out to make deliveries.
Amazon was one of the first to explore humanoid robotics at scale when it began testing Digit by Agility Robotics in 2023. Oregon-based Agility has been mentioned as one of the companies Amazon is considering for its last mile ambitions.