Machenta exits stealth with $40M seed: sim-only foundation model teaches robots from one demo
Ed__Johns · x · 2026-10-27
Machenta, a London startup founded by Ed Johns, head of the Robot Learning Lab at Imperial College, exits stealth with a $40M seed round co-led by Creandum and USV.
Key technical points:
- Vision-demo-action model: like a VLA, but the task is given as a demonstration instead of a sentence; demo plus live camera images flow through the model, which outputs actions in a closed loop with no weight changes at test time
- Trained entirely in simulation on randomized objects, tasks, scenes and robots, so the weights learn how to pick up a task from a demo rather than any one task — zero real-world training data
- Traditional robot learning takes hours/days of data collection and training per task; Machenta removes all post-training
The site shows reactivity and recovery, real industrial tasks, a one-person one-shot factory, and humanoids — all learned instantly by robots that have never seen the objects. The team calls the experience "physical vibe coding" and plans to triple down on the Bitter Lesson.
Related event: Robotics Startup Machenta Exits Stealth with $40M Seed Round(16 posts)→
More from Venture
- The real AI risk to Indian IT: every dollar earned comes with an OpenAI subscription — NirantK · 2026-10-27
- Elvis estate sues Anthropic over 274 songs as LSEG sells data for $3.535B — YvesMulkers · 2026-10-27
- Hyphenate raises $45M Series A as revenue grows 86x with autonomous finance agents — aakashgupta · 2026-10-27
- Toyota Pours $6.4B a Year Into Humanoid Robots, Targets 400,000 by 2028 — emmanuelvivier · 2026-10-27
- TrustMRR racks up 98 mentions in 24 hours and even Marc Lou isn't sure why — marclou · 2026-10-27
- Indie dev ports taliri to Firefox-based browsers in a week after a $49 refund — theteknosaur · 2026-10-27