瑞士机器人初创公司 Enact Intelligence 获 3000 万美元 pre-seed 融资

Swiss robotics startup Enact Intelligence today announced that it has raised $30 million in pre-seed funding.

The capital was provided by a consortium that included Yann LeCun, one of the world’s most prominent artificial intelligence researchers. He was joined by Mistral AI SAS Chief Executive Arthur Mensch, Hugging Face Inc. co-founder Thomas Wolf and several venture capital firms.

Enact, officially Eta Robotics AG, was launched earlier this year by CEO Manuel Yves Galliker. The executive earlier held engineering roles at robotics startups Physical Intelligence Inc. and 1X Technologies AS. The Enact team also includes former employees of Google DeepMind, Amazon.com Inc. and other major industry players.

Enact is developing AI models optimized to control robots. Such algorithms were historically tailored to specific robot form factors. As a result, a model built to power a humanoid robot may not be capable of reliably operating a standalone robotic arm. Enact’s algorithms, in contrast, are designed to support a wide range of robot designs.

Early-stage startups often use customized versions of open-source models to power their software. A recent job posting indicates that Enact is pursuing a different approach. The company is hiring an AI researcher with expertise in pre-training, a term for developing AI models from scratch. That task can be significantly more difficult than customizing open-source models.

Enact says that it’s developing not only AI models but also physical robot components, most notably actuators. Those are mechanical subsystems that act as a robot’s muscles. An actuator combines an electric motor with a chip that controls its motion and certain other auxiliary parts.

Enact’s actuators not only produce kinetic energy but also act as sensors. It says they “sense force and touch,” which presumably means they measure the amount of pressure a robot applies to objects when it picks them up.

The company didn’t specify how the technology works. However, earlier attempts to build actuator-based sensors may offer clues. Google LLC and publicly traded chipmaker Allegro MicroSystems Inc. are among the companies that have invested in such projects. Both companies used an approach called Back EMF to turn actuators into pressure sensors.

Running a current through an actuator causes its built-in motor to spin. The rotation of that motor, in turn, generates a second current that flows back into the actuator. Back EMF technologies measure that second current, or counter-electromotive force as it’s most commonly known, to understand what the host system is doing.

If a robotic arm’s actuator is active but not generating any counter-electromotive force, engineers can deduce that the arm is holding a heavy item its actuator is unable to move. By contrast, the presence of a strong counter-electromotive force indicates that the actuator is spinning freely. That often means the robot arm isn’t holding anything. Back EMF sensors can also collect data about a range of other robot activities.

A job posting indicates that Enact plans to use field-programmable gate arrays, or FPGAs, to power its robots. An FPGA is a chip that users can customize more extensively than a central processing unit or a graphics card. That makes it possible to achieve better performance and power efficiency.

Enact will reportedly use its funding round to hire more engineers and enhance its technology.

Photo: Unsplash