The company called the program Tasks, which lists digital assignments couriers can accept in place of or alongside standard delivery orders. Tasks range from recording unscripted conversations in Spanish to filming household activities such as loading a dishwasher, handwashing dishes and folding clothes.
Instructions for the dishwashing task require a body-worn camera pointed at the worker’s hands, scrubbing and rinsing at least five dishes, and holding each clean dish steady in frame before moving to the next, according to Bloomberg. Robotics firms use that footage to train humanoid systems to recognize objects and execute contact-rich manipulation tasks.
From Package Drops to Data Collection
The scope of what DoorDash is building extends beyond household video. Through the company’s regular courier app, it deploys workers for in-field data capture at commercial locations. Couriers can accept assignments to scan supermarket shelves for inventory checks, photograph hotel entrances to tag drop-off locations, or capture food images to populate restaurant digital menus.
A DoorDash spokesperson told Bloomberg that the company uses the submitted audio and video footage to evaluate both in-house AI models and those built by partners across retail, insurance, hospitality and technology sectors.
The program currently excludes heavily regulated markets, including California, New York City, Seattle, and Colorado. DoorDash said it plans to expand the types of tasks and its geographic coverage over time. The company’s pilot with Alphabet’s Waymo, in which drivers close robotaxi doors for pay, also falls under the Tasks umbrella, placing autonomous vehicle training alongside shelf scanning and household footage as parallel outputs of the same distributed workforce.
Advertisement: Scroll to Continue
“These are the kinds of real-world problems we’ve been solving for over a decade, and we realized the same capabilities that helped us could help other businesses too,” Ethan Beatty, general manager of DoorDash Tasks, said in a press release. “The goal of Tasks is to help more businesses understand what’s happening on the ground and gather new insights, all while giving Dashers a new way to earn.”
Uber and the Emerging Data Labor Market
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
DoorDash is not alone in this shift. Uber introduced a comparable program in October, adding a digital tasks category to its driver app that allows registered drivers to complete short assignments such as uploading restaurant menus and recording multilingual audio samples.
The effort operates through Uber AI Solutions, the company’s enterprise data services division, which has expanded to 30 countries and offers annotation, translation and model training services to corporate clients, PYMNTS reported. Uber also acquired Segments.ai, a lidar and multi-sensor annotation startup, to deepen its capabilities in perception data for robotics and autonomous systems.
Both companies follow a path pioneered by data infrastructure firms like Scale AI, using distributed networks of remote workers to create new datasets or validate AI outputs. What gig platforms add to this model is scale, geographic reach and access to the physical world in its most variable, uncontrolled form.
Gig Platforms as AI Training Infrastructure
That access is precisely what makes this moment significant. Physical AI systems, including humanoid robots, autonomous vehicles and warehouse automation, cannot be trained solely on clean simulations.
Universal Robots and Scale AI made this case directly in an announcement on Monday (March 16), unveiling an imitation learning system designed to capture high-fidelity, synchronized robot and vision data in production environments.
Anders Beck, vice president of AI robotics oroducts at Universal Robots, said in the announcement that most training data collected on research robots is not suited for real-world deployment, and that visual feedback alone fails for contact-rich tasks. The gap between lab and factory performance remains one of the central unsolved problems in physical AI, and closing it requires data from real-world environments.
The longer-term implication is that real-world data collected at scale from distributed human workers is becoming a meaningful competitive asset. Platforms with large contractor bases, established presence in physical commercial environments and logistics infrastructure to coordinate task-based workflows are positioned to accumulate proprietary training datasets that AI developers and robotics firms cannot easily replicate.
For all PYMNTS AI and digital transformation coverage, subscribe to the daily AI and Digital Transformation Newsletters.
Add as Preferred Source
