Household robotics is entering its "GPT moment." Sunday Robotics has announced a massive $165 million Series B funding round to commercialize its autonomous home assistants, leveraging a unique "Skill Capture" methodology that sidesteps the data bottlenecks of traditional robotics.
While robots have excelled in factories for decades, the unstructured environment of a typical home—with its messy counters, varied lighting, and unpredictable pets—has remained an unsolved challenge. Traditional training requires millions of hours of simulation, which often fails to capture the "feel" of delicate tasks like handling eggs or folding silk.
Sunday Robotics' breakthrough is the Skill Capture Glove. Instead of training via reinforcement learning in a vacuum, human "Teachers" wear these gloves while performing tasks. The gloves capture 1,000 data points per second, including pressure, orientation, and subtle muscular adjustments. This proprioceptive data is then fed into a transformer-based model, allowing the robot to "feel" the physics of the task before it ever attempts it.
The robot doesn't just copy movements; it understands intent. By integrating a multi-modal VLM, a user can say, "Clean up the red wine spill but don't use the good napkins." The model translates this natural language into a series of visual sub-goals, identifies the spill, locates a rag, and executes the physical movements captured via the Teacher sessions.
Automate your physical and digital tasks with our professional agent development suite.
In pilot tests conducted across 500 varied household layouts, Sunday's "Sunday v2" hardware achieved staggering results:
Technically, Sunday Robotics utilizes what they call a Proprioceptive Transformer (PT). Unlike standard LLMs that predict the next token, the PT predicts the next set of motor torques based on a window of historical tactile feedback and current visual input. This creates a closed-loop system that can adjust in real-time if a glass slips or a table is bumped.
Sunday Robotics isn't just building a robot; they are building a Physical OS. With $165M in the bank, the goal is to drive down hardware costs to under $20,000 per unit, making general-purpose household help a reality for the upper-middle class by late 2027. The era of the "Robot-as-a-Service" for chores has officially begun.
For more on how AI is interacting with the physical world, read about the Tesla Optimus Gen 3.