Ep#74: Weave Robotics

In this talk

1. 0:51

   Designing a useful home robot

   Khan says Weave started because the team saw room for a practical home robot sooner than five to ten years out. Its early prototype used wheels for long operating hours, a collapsible body to stay out of the way, and arms that could reach desks and high cabinets. The team tested its design by living with the robot.

2. 3:19

   Choosing laundry folding as a first task

   Weave focused on laundry folding and tidying, then found folding could be useful as a defined task customers could run while the robot stayed in place. Tidying remains a goal, but Khan says it is hard to scope: requests about keys, shoes, toys, and plants quickly create many possible behaviors. The team says customers use its folding service every day.

3. 5:43

   Building hardware and learning from deployments

   Khan says off-the-shelf arms struggled with the payload, reach, speed, and fabric handling needed for garments such as heavy pants and large towels. Weave iterated its own hardware, including five revisions of its gripper. Deployed robots run an autonomy stack, with progress models detecting when they are stuck or undoing work so a person can intervene.

4. 12:54

   Making models adaptable and collecting useful data

   Customer preferences for how shirts should be folded showed Weave that models need to adapt to different protocols, through prompting, fine-tuning, or other methods. Khan also stresses that more data alone cannot fix a poorly designed task strategy, using the different handling needs of short-sleeved and long-sleeved shirts as an example. The team favors autonomous rollouts with targeted interventions to gather examples of difficult subtasks.

5. 22:13

   Selling a product and choosing the next steps

   Weave charges for its robots: home customers can pay upfront or subscribe, while business pricing has also been flat. Isaac Zero is a stationary version of a mobile robot the company is developing; Bay Area customers can order one with in-home deployment, and the team offers local repair support. Khan says the company wants to prove it can do laundry well before making bigger claims about future tasks.