Sunday Robotics secures $165M Series B to bring autonomous robots to homes using Skill Capture Glove and Vision-Language Models.
What the Series B Is Really Funding
Sunday Robotics has raised a $165M Series B aimed at putting autonomous robots in homes. At that stage, capital usually goes less toward proof-of-concept demos and more toward the hard parts of home deployment: reliable manipulation, safe operation around people and pets, and the data loop that turns rare edge cases into recoverable skills. Households are messier than warehouses—cluttered counters, variable lighting, soft objects, and tasks that change every day—so progress depends on systems that learn from real human motion rather than only from scripted routines.
The headline number matters mainly as a signal of scale. Building and supporting home robots requires hardware iteration, fleet software, and continuous model updates. A large Series B is a bet that those costs can be amortized if the same skill-learning stack works across many chores and many floor plans instead of one narrow task in one lab kitchen.
Skill Capture Gloves as a Training Interface
The Skill Capture Glove is a practical answer to a core robotics bottleneck: how do you teach a machine a household skill without writing a brittle program for every motion? A glove worn by a person can record hand pose, contact timing, and force-related cues while someone folds laundry, loads a dishwasher, or clears a table. Those trajectories become training data for policies that map what the robot sees and feels to what it should do next.
That approach has clear tradeoffs. Human demonstrations capture nuance that pure simulation often misses—how people re-grip a slippery plate, or pause when a child walks through the room. They also introduce bias: every demonstrator has habits, and every glove setup has sensing limits. Teams that use skill capture well treat demos as seeds, then refine policies with correction, failure recovery, and multi-user variation so the robot is not locked to one person’s style.
Vision-Language Models in the Home Stack
Vision-Language Models (VLMs) give the robot a way to ground open-ended instructions in pixels. Instead of only recognizing a fixed set of objects, a VLM can help interpret “put the mugs by the sink” or “stack the dry towels on the shelf” by linking language to scenes the robot has never seen in that exact layout. That is essential in homes, where object sets and room layouts differ by household.
VLMs alone do not solve control. They reduce ambiguity about what the task is and where relevant objects are; low-level policies still have to reach, grasp, and place without breaking things. A workable architecture usually keeps language understanding, skill libraries from glove demos, and safety constraints as separate layers: the VLM proposes intent and candidates, the skill system executes, and a safety layer vetoes unsafe motions.
- Use language to select and sequence skills, not to micromanage every joint angle.
- Fall back to clarification or a safe stop when the scene does not match the instruction.
- Log failures with video and language so the next skill-capture session targets real gaps.
What “Autonomous at Home” Has to Mean in Practice
For Sunday Robotics’ goal—autonomous robots in homes—autonomy is less about never asking for help and more about finishing common chores with minimal supervision. That means graceful recovery when a grasp fails, clear user overrides, and predictable behavior near stairs, pets, and open cabinets. Shipping a product in this category is as much about trust and maintainability as about raw capability: owners need to understand what the robot can do today, how new skills get added, and when human intervention is expected.
If skill capture and VLMs work together as intended, new household tasks can enter the system as demos plus language descriptions rather than months of hand-coded routines. The $165M Series B is the resource pool for proving that loop at home scale—where reliability, not a single impressive video, decides whether autonomous household robots stick.