Researchers and startups are training robots to perform household chores using video recordings of humans folding laundry, a method that could reshape domestic robotics.
The approach leverages machine learning algorithms that analyze human motion captured through head-mounted smartphones, translating those movements into precise robotic actions for tasks like folding clothes or making beds.
DoorDash has joined a growing network of companies paying gig workers up to $25 an hour to record themselves performing chores, creating a dataset essential for teaching robots complex physical skills.
This data collection effort mirrors the early training of AI chatbots and image generators, which relied on vast libraries of text and photos to achieve human-like outputs.
To bridge the gap between human dexterity and robotic motion, researchers also film themselves manipulating robotic limbs while folding clothes, providing direct examples of how machines should replicate human techniques.
Recordings span diverse global participants, different clothing types, and varied surfaces to ensure the AI can generalize across real-world household conditions.
Once trained, the AI model is embedded into robots that attempt autonomous folding, using camera and sensor input to predict and execute the next movement in real time.
Meanwhile, a separate innovation from a Palo Alto startup demonstrates how such technology can be integrated into everyday furniture, with a floor lamp that unfolds robotic arms to tidy bedding before retracting into its original form.
The Lume device, developed by engineers Aaron Tan and Angus Fung, originated from their frustration with bulky home robots that clashed with domestic spaces, prompting a design that hides functionality within familiar objects.
Inspired by observing how appliances like washers and dryers blend into rooms until needed, the creators aimed for a robot that remains unobtrusive until activated by user command.
Lume’s mechanism includes soft fabric-covered joints to prevent entanglement, motion-sensing safety systems that halt operation near children or pets, and motors that slow when non-fabric objects are detected.
As the device remains stationary, it avoids mechanical wear from movement across floors or doorways, a common failure point in mobile home robots.
The system’s training relied on hours of manually recorded folding footage, which was processed to replicate gestures thousands of times until the motion appeared fluid and natural.
How does using video of people folding laundry improve robot training compared to traditional programming?
It captures the nuanced, adaptive motions of human hands that are difficult to encode through rigid code, allowing robots to learn variability in fabric types, folding techniques, and real-world conditions.
What safety features distinguish the Lume robot lamp from earlier home robot prototypes?
It uses pressure-sensitive joints with fabric covers, environmental sensors that stop operation near people or pets, and torque-limiting motors that reduce force when encountering unexpected objects.