A robot in a yellow safety vest takes a parcel from the bed of a pickup and walks towards a building. A little later, a humanoid picks up a box on some steps. The two-legged delivery robot has a very familiar job.
The footage comes from Reward AI, which demonstrated its OM-1 AI model in several robots on 27 October. The company develops software that teaches these machines to act. Carrying parcels is one application; in the same demonstration, a robot also cleans a floor and opens a fridge.
For a delivery robot, the route does not end at the kerb. Taking a parcel out of a vehicle, holding it while walking and putting it in the right place are separate actions that all have to work together. That combination makes this demonstration interesting: the robot moves around and uses its hands.
Reward AI describes the demonstrated actions as fully autonomous and says the footage runs at normal speed. This is a developer demonstration, rather than an announced parcel service with regular routes and customers.
Learning from human hands
What stands out is how the robots learn. In its technical explanation of OM-1, Reward AI describes a wearable device called Omnibody Hand. People wear it while performing tasks. Cameras and sensors record what they see, touch and grasp, how their hands move and how much force they apply.
This lets the company gather examples of human work instead of remotely controlling a robot through every movement. The AI model learns which action should follow. The approach is intended for different machines, from robot arms to humanoids.
A separate control system underneath translates movements into commands for each robot’s motors. That layer is trained in a computer simulation. It needs to keep making adjustments when, for example, a box proves heavier than expected or a fridge door resists.
Reward AI says OM-1 can learn a new task from less than thirty minutes of examples. That is a claim about training data, not a promise that a robot can safely deliver anywhere after half an hour.
When will it reach your doorstep?
When might such a robot arrive with your post or meal? Reward AI has not given a start date for this application. The footage shows part of the work, but not how many consecutive deliveries succeed, what a delivery costs or how the robot handles an unfamiliar address. Those questions matter for a real delivery service.

























