"Instant Robot Learning."
London Imperial College spinout incorporated 2025 building a sim-trained robot model that learns tasks from one demonstration; raised a $40M seed round.
Machenta is a London robot-learning company spun out of the Robot Learning Lab at Imperial College London, developing a robot foundation model that learns a new manipulation task from a single demonstration with no further training. Machenta Ltd was incorporated in England and Wales on 21 July 2025 (company number 16594535), with a registered office on Elthorne Road, London N19, and the team works in person in King's Cross. Edward Johns, Professor of Robot Learning at Imperial, is co-founder and chief executive. Vitalis Vosylius, whose Imperial PhD produced the Implicit Graph Alignment (CoRL 2023) and Instant Policy (ICLR 2025) in-context learning models, is co-founder and chief technology officer.
The company's approach, which it calls Instant Robot Learning, builds on six years of one-shot learning research at Imperial, including Learning a Thousand Tasks in a Day, which appeared on the cover of Science Robotics. Instead of a vision-language-action model that is post-trained on many demonstrations per task, Machenta trains what it calls a vision-demo-action (VDA) model: the task is specified by one demonstration rather than a text prompt, and the network is meta-trained to learn how to learn new tasks. Pre-training uses only simulated data, generated at scale so that capability grows with compute; real-world data is used only for the single teaching demonstration at deployment.
In material published in October 2026, the company showed the same pre-trained model running on six different robots with different grippers and camera layouts, including humanoids and low-cost hobbyist arms as well as industrial arms, taught through four modes: leader-follower teleoperation, VR controllers, kinaesthetic teaching and the teacher's own hands. Demonstrated tasks include toolbox and workshop tidying, stacking cans in a rack, cable sorting, paper roll replacement, tending a bag sealer and a centrifuge, and mannequin dressing, with the robot handling objects it had not seen before and recovering from disturbances. One video shows three robots configured as a small assembly line within two minutes. Extra demonstrations can be added to clarify ambiguous tasks, combine object attributes or stitch long-horizon tasks together from sub-demonstrations. The model is not yet generally available; Machenta runs an early-access registration that asks which robot arms prospective users plan to use.
In October 2026 Machenta announced a 40 million US dollar seed round co-led by Creandum and Union Square Ventures, which Imperial College described as the largest seed financing for a UK robotics startup. The company said the money will scale the foundation model and expand operations at a new King's Cross facility, and it is hiring in robot learning, robot control, simulation and 3D machine learning.
Primary type & automation activities this supplier delivers:
Sim-trained vision-demo-action foundation model that learns a manipulation task from one demonstration, no fine-tuning.
Contact Machenta
WEBSITE
https://machenta.comHEADQUARTERS
United Kingdom
Company Facts
Founded
2025
Primary Role
Software/Algorithm
Company Size
-
Primary Region
Europe
Annual Sales
-
Funding Stage
Early-Stage VC
Funding Total
-