News Summaries

Robots Need Data Too: Startup XDOF Pays Labs to Do the Grimy Training Work

Training a robot is a lot messier than training a language model. Collecting the real‑world footage that lets a robot learn to pick up objects, twist knobs or hand over tools is tedious, expensive and requires a lot of hardware. That’s the problem a new startup, XDOF, is turning into a business. Founded in October 2024 by robotics veterans Wu, Shentu and Jin, XDOF offers a full‑service data pipeline for companies building robot‑learning models. The name plays on “degrees of freedom,” the many ways a robot arm can move, and the firm promises to handle “arbitrary” motion data at any scale. XDOF’s first big win is a partnership with UC Berkeley’s AI lab to release what they call the ABC dataset – the largest collection of high‑quality robot‑training data ever assembled. It contains 130,000 manipulation trajectories, 300 hours of simulated runs and 100 hours of real‑world evaluations. The data isn’t just dumped on researchers; XDOF also cleans, annotates and builds tools around it, creating a feedback loop that keeps the dataset fresh and useful. By outsourcing this “dirty, unglamorous” work, AI labs can focus on model design while XDOF handles the heavy lifting, much like how open‑source image and language datasets sparked rapid breakthroughs in those fields. The company hopes this model will become the go‑to source for robot‑learning data, accelerating the pace of robotic innovation.

Secret Quakes Under Antarctica Reveal a Surprising Hotspot Beneath the Ice

A team of geophysicists has uncovered a hidden swarm of tiny earthquakes deep beneath Antarctica’s ice sheet, shaking up the long‑held belief that the frozen continent is seismically quiet. By re‑examining seismic recordings collected between 2001 and 2004 with modern analysis tools, researchers identified hundreds of low‑magnitude tremors clustered in an unexpected region far from any known fault lines. "Antarctica was thought to lack earthquakes, but that was just because we didn’t have the right ears," said Richard Alley, a glaciologist at Penn State who was not involved in the study. The newly detected quakes are too weak to threaten the massive ice shelves or the fragile ecosystem above, but their presence tells scientists that the continent is more dynamic than previously imagined. The findings suggest hidden stresses in the Earth’s crust beneath the ice, possibly linked to the slow movement of the underlying rock or the weight of the ice itself. While the tremors pose no immediate danger, they open a fresh window into Antarctica’s geological activity and may help improve models of ice‑sheet behavior in a warming world. The discovery underscores how revisiting old data with fresh techniques can reveal surprises lurking beneath even the most remote corners of our planet.

11 YC Demo‑Day Startups VCs Can’t Stop Talking About

Y Combinator’s latest Demo Day showcased a handful of companies that have captured the imagination of venture capitalists. First up is Arga Labs, which tackles a growing bottleneck for developers: testing AI‑generated code. Its platform instantly creates “digital twins” of a company’s software stack, giving AI agents a safe sandbox to run tests before code hits production. Adialante is another eye‑catcher, rolling out mobile MRI units that bring early‑cancer screening directly to patients’ neighborhoods, promising faster diagnoses and lower costs. Tasklet rounds out the list with a conversational AI assistant that plugs into everyday work tools—Slack, Outlook, Google Drive, and more. Users can simply ask the agent to sort emails, pull reports, or even write and execute code, and it keeps working in the background even after the browser tab closes. The event, held in Los Angeles on June 18, also featured fireside chats with leaders from Mach Industries, Founders Fund, and Shinkei Systems, offering founders practical advice on scaling their businesses. Across the board, these startups illustrate how AI, mobile health, and seamless automation are reshaping the tech landscape, and why VCs are eager to back them.