ENACT raises nearly 27 million euros to develop tactile robotics

ENACT, a Swiss startup, has raised nearly 27 million euros in pre‑seed funding to develop robots that can better perceive forces and adjust their movements when they touch objects. The company plans to design together the mechanics, data and artificial intelligence models needed for manipulation. In France, Yann LeCun and Arthur Mensch are participating in the financing. Manipulating a cable, using a tool, performing an assembly are part of Enact Intelligence's data collection program. These gestures require the robot to interact with matter, accompany a movement and adapt to what happens during the action. For the Swiss company, part of the solution lies in the information the machine can gather about its own movements and the forces it encounters. Enact Intelligence aims to develop robotics capable of learning fine manipulation by combining AI models, sensors and actuators, the components that produce motion. The question concerns both the design of the robot and the design of its learning: what information must be provided for it to adjust its gesture correctly? Learning what happens at contact The startup describes its approach as a proprioceptive system linked to models that can exploit force information. Proprioception informs the robot about its own mechanical state; perception of effort must help it adapt its behavior during manipulation. The "sense of touch" here covers the ability to take into account physical interaction. Enact emphasizes forces and proprioception, without detailing the full sensor architecture, it would be premature to imagine an artificial skin reproducing all the sensations of a human hand. The goal is to enable the learning of manipulation skills usable on different tasks and in different environments.

An ambition discovered in the analysis of the startup's recruitment: future scientists must work on force‑sensitive models, organize training data and evaluate behaviors on real robots. The program includes the efficiency of learning and robustness when moving to real conditions. The four links of the project Sensors collect the information needed to estimate the robot's state and its interactions. Actuators produce the movements that the system must control. Human demonstrations provide examples of gestures to learn. AI models are trained and evaluated to transform this information into robotic behaviors. Enact seeks to develop these links together. The gain obtained on an industrial task remains to be established by measured results. The mechanics enters the learning loop Manuel Galliker, CEO and founder of the company, previously worked in robotic control at 1X. He wants to rethink the data loop and how the robot's body, through its actuators and sensors, can provide additional information to learn fine manipulation. Enact is looking for skills in mechanical design, programmable electronics, real‑time control and integration of systems. The company also wants to develop the infrastructure needed to train and evaluate its models. This integration can allow the hardware and learning methods to evolve together. The company commits to several fronts: a better model must work with reliable components, and mechanical changes must remain compatible with the data and software used. The breadth of the program explains the projects the financing must address, so the startup must fund AI experiments, manufacture and test hardware, and gather the skills capable of making the whole system work. Before autonomy, human gestures to record One of the most revealing positions among those to be filled is that of robot operator.

Its mission is to guide the machines and teleoperate them using motion‑capture systems. These demonstrations must directly feed the training of neural networks. The work includes quality control of the data, annotation of trajectories and reporting of unusual behaviors. Enact's program therefore provides a structured activity of producing physical examples. This organization places the efficiency of data collection at the center of the project. How many demonstrations are needed to learn an operation? What share of this learning can be reused elsewhere? At what frequency must an operator take back control? Answers that will determine part of the cost of adapting the system to an industrial need. In Zurich, different approaches to dexterity confront each other Enact joins a field already occupied by mimic Robotics. The other Swiss company develops an integrated approach linking a robotic hand, learning models and devices for collecting human demonstrations. Other companies address the problem with models capable of commanding different robots. Physical Intelligence presented a general model trained on several platforms, while Flexion developed a layer of autonomy intended for various hardware. These approaches raise a common question: what share of intelligence can be transferred between machines, and what share depends on their physical design? The industrial choice will also depend on existing solutions: specialized automation, custom integration or manual work. The next milestone is measured in duration Enact is also preparing criteria

Reported by FrenchWeb.