Several robotics firms are addressing one of AI's toughest challenges: creating foundation models that handle tasks without prior specific training. Their methods vary widely, ranging from analyzing vast numbers of online videos and running computer simulations to gathering motion information from people doing activities while wearing sensor-equipped gloves.

Enigma, a research lab preparing to exit stealth mode on Monday, pursues a distinct strategy. Instead of centering solely on model performance, the startup under one year old aims to examine human-robot engagement to develop natural interfaces and potentially an alternative form of robotic intelligence.

To support its goals, Enigma obtained a $70 million seed round led by Index Ventures and Ribbit Capital, with involvement from Sarah Guo of Conviction Partners.

To explore human preferences in machine communication, Enigma is starting a broad trial that lets anyone worldwide engage remotely with over 100 of its custom AI robots. These machines, kept in facilities in Israel and California, can execute actions like sketching with a brush, dueling via swords, and conducting basic chemistry by handling and combining liquid-filled containers. The company states it built both the arms and their core models entirely in-house.

Jonathan Jacobi, Microsoft's youngest-ever hire recruited by Wiz founder Asaf Rappaport, established Enigma alongside longtime friend Gal Niv. The pair connected through teen hacking contests and grew close during service in Israel's Unit 8200, focusing on cybersecurity studies.

When Jacobi and Niv launched their venture last year, they chose to direct their skills toward AI robotics despite limited prior background, viewing it as the area with the most compelling open questions in technology.

They built a group of what Jacobi calls some of their brightest contacts from Israel's tech scene, featuring former members of leading AI groups, math competition winners, and several individuals persuaded to leave PhD studies.

Shardul Shah, partner at Index Ventures, noted that while many robotics veterans join the embodied intelligence push, Jacobi and Niv bring an outsider perspective that allows greater originality. Insiders might begin with teleoperation or dexterity, yet Enigma starts instead with the ideal user experience.

Jacobi explained to TechCrunch that Enigma seeks to render human-robot exchanges entirely seamless. He described how users facing a 15-minute explanation for a simple chore like dish placement often give up and handle it themselves, a frustration common even with advanced models today.

Jacobi argues that robot control should feel as natural as adjusting a car stereo volume. He points out the annoyance of specifying exact percentages without clear feedback on the outcome.

Enigma intends for insights from its online trial to identify an interface akin to that volume control. The public test will assess various methods for directing its robots.

Jacobi noted the experiment will clarify preferred interaction styles, such as text or audio commands, video demonstrations, or drag-and-drop gestures.

He acknowledged the trial's broad scope, with the aim of collecting interaction data to improve both interfaces and model training approaches.

While the company may identify optimal communication methods, its commercial applications are still developing. Jacobi mentioned partnerships already underway in healthcare, logistics, and entertainment without detailing specific uses.