# Machenta raises $40M seed to cut robot training time with one demonstration

Digest AI · Robotics & Physical AI · published 2026-10-27T13:58:12Z · updated 2026-10-27T14:35:01Z

Canonical: https://digestai.news/story/machenta-raises-40m-seed-to-cut-robot-training-time-with-one-demonstra

## Summary

Machenta, a London-based robotics startup, announced a $40 million seed round on October 27, co-led by Creandum and Union Square Ventures. The company aims to reduce the time and effort needed to teach robots new tasks by enabling them to learn from a single demonstration, without additional training data collection or model updates. Its approach, called *instant robot learning*, targets workflows like packaging, assembly, and tidying, where robots adapt to variations after seeing the task once. The technology relies on simulation for pretraining to build a reusable learning ability, then uses real-world demonstrations and observations at deployment time to execute tasks in real environments.

Machenta’s co-founder and CEO, Professor Edward Johns, and CTO Vitalis Vosylius, built the work on research from Imperial College London, including the 2025 *Instant Policy* paper on in-context imitation learning. Early trials show high success rates—74 out of 74 block placements, 78 out of 78 screwdriver handovers, and 97 out of 99 bottle-standing attempts—but the company acknowledges limitations in real-world reliability, safety, and hardware constraints. The funding will support scaling the model and expanding its team at a new lab in King’s Cross, London. Investors highlight the potential to cut training bottlenecks, making robots more practical for dynamic tasks in factories and smaller-scale applications.

## Key points

- Machenta raised $40M seed round co-led by Creandum and Union Square Ventures on October 27
- Startup’s model learns physical tasks from one demonstration without retraining, targeting packaging, assembly, and tidying
- Early trials show 97% success on bottle-standing tasks but do not prove broad industrial reliability

## Why it matters

If robots can learn tasks from a single demonstration, companies could deploy them for more flexible workflows without lengthy reprogramming.

## Sources

1. [Machenta Raises $40M Seed to Teach Robots New Tasks From One Demonstration](https://unite.ai/machenta-raises-40m-seed-instant-robot-learning) (Unite.AI, 2026-10-27)
2. [London-based Machenta, which is building an AI model that can teach robots new physical tasks after seeing them performed just once, raised a $40M seed](https://businessinsider.com/this-11-person-startup-raised-40-million-train-robots-virtual-worlds-2026-10) (businessinsider.com, 2026-10-27)

## Cite

Digest AI, "Machenta raises $40M seed to cut robot training time with one demonstration", 27 October 2026, https://digestai.news/story/machenta-raises-40m-seed-to-cut-robot-training-time-with-one-demonstra

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/machenta-raises-40m-seed-to-cut-robot-training-time-with-one-demonstra.json