Cheap teleop gloves scale humanoid data capture
What happened
Sunday Robotics is pushing low‑cost 'Skill Capture Gloves' (~$200 vs $20k teleop rigs) to scale human chore demonstrations into robot data, claiming 90%+ success when combined with Skill Transform tools for foundation‑model style skill transfer noted. The move could materially lower the cost of collecting broad manipulation datasets for embodied agents.
Why it matters
Sunday closed an oversubscribed $165M Series B led by Coatue at about a $1.15B valuation, with Coatue partner Thomas Laffont joining Sunday’s board. (sunday.ai) The company says its training corpus contains roughly 10 million in‑home task episodes collected from more than 500 households, material the team used to build a robotics foundation model (ACT‑1). (aibusiness.com) Sunday reports it has shipped over 2,000 Skill Capture Gloves to “Memory Developers” in the field to gather that data at scale. (originofbots.com) The glove concept was iterated heavily in hardware and form factor — Sunday’s team has said they ran nearly 100 design iterations on the wearable (branded UMI, for Universal Manipulator Interface) before finalizing the product. (humanoidsdaily.com) Sunday describes Memo’s training set as enabling multi‑step home tasks such as loading dishwashers, folding laundry, clearing tables and brewing coffee, according to their product materials and demonstrations. (humanoidroboticstechnology.com) To move from lab demos to households, Sunday says it tripled engineering headcount, quadrupled research, and grew data operations fivefold ahead of a beta rollout the company plans to start in the fall of 2026. (sunday.ai) The startup’s leadership traces to Stanford robotics research and prior stints at DeepMind, Tesla and GoogleX for co‑founder Tony Zhao, with co‑founder Cheng Chi holding academic appointments at Stanford and Columbia. (tonyzhaozh.github.io)
Key numbers
- Sunday Robotics is pushing low‑cost 'Skill Capture Gloves' (~$200 vs $20k teleop rigs) to scale human chore demonstrations into robot data, claiming 90%+ success when combined with Skill Transform tools for foundation‑model style skill transfer noted.
- Sunday closed an oversubscribed $165M Series B led by Coatue at about a $1.15B valuation, with Coatue partner Thomas Laffont joining Sunday’s board.
- (sunday.ai) The company says its training corpus contains roughly 10 million in‑home task episodes collected from more than 500 households, material the team used to build a robotics foundation model (ACT‑1).
- (aibusiness.com) Sunday reports it has shipped over 2,000 Skill Capture Gloves to “Memory Developers” in the field to gather that data at scale.
What happens next
- (humanoidroboticstechnology.com) To move from lab demos to households, Sunday says it tripled engineering headcount, quadrupled research, and grew data operations fivefold ahead of a beta rollout the company plans to start in the fall of 2026.
- The move could materially lower the cost of collecting broad manipulation datasets for embodied agents.
Quick answers
What happened in Cheap teleop gloves scale humanoid data capture?
Sunday Robotics is pushing low‑cost 'Skill Capture Gloves' (~$200 vs $20k teleop rigs) to scale human chore demonstrations into robot data, claiming 90%+ success when combined with Skill Transform tools for foundation‑model style skill transfer noted. The move could materially lower the cost of collecting broad manipulation datasets for embodied agents.
Why does Cheap teleop gloves scale humanoid data capture matter?
Sunday closed an oversubscribed $165M Series B led by Coatue at about a $1.15B valuation, with Coatue partner Thomas Laffont joining Sunday’s board. (sunday.ai) The company says its training corpus contains roughly 10 million in‑home task episodes collected from more than 500 households, material the team used to build a robotics foundation model (ACT‑1). (aibusiness.com) Sunday reports it has shipped over 2,000 Skill Capture Gloves to “Memory Developers” in the field to gather that data at scale. (originofbots.com) The glove concept was iterated heavily in hardware and form factor — Sunday’s team has said they ran nearly 100 design iterations on the wearable (branded UMI, for Universal Manipulator Interface) before finalizing the product. (humanoidsdaily.com) Sunday describes Memo’s training set as enabling multi‑step home tasks such as loading dishwashers, folding laundry, clearing tables and brewing coffee, according to their product materials and demonstrations. (humanoidroboticstechnology.com) To move from lab demos to households, Sunday says it tripled engineering headcount, quadrupled research, and grew data operations fivefold ahead of a beta rollout the company plans to start in the fall of 2026. (sunday.ai) The startup’s leadership traces to Stanford robotics research and prior stints at DeepMind, Tesla and GoogleX for co‑founder Tony Zhao, with co‑founder Cheng Chi holding academic appointments at Stanford and Columbia. (tonyzhaozh.github.io)
