Samsung Galaxy Ring wins FDA clearance for AI sleep apnea grading

Samsung announced on 3 November that it upgraded the Galaxy Ring's sleep apnea detection, after the FDA approved the new feature in July, Engadget reported. The wearable now uses AI to count breathing disruptions per hour of sleep.

The original February 2024 De Novo authorisation covered only symptoms of moderate to severe apnea. The new version adds mild, occasional and frequent categories. It requires the wearer to log at least five hours of sleep on two nights within a ten day window. That is a device clearance, not a road ruling. But the same week produced a cluster of decisions that matter more for self-driving regulation. They point in different directions.

Approval, then a court, then a vote

On 2 November, shareholders of Churchill Capital Corp XI voted to approve the SPAC merger with Agility Robotics, according to The Robot Report. More than 84% of votes cast at a special meeting were in favour. The deal is expected to close on 4 November, with trading on Nasdaq starting 5 November under the ticker AGLT. The transaction is expected to deliver roughly $465 million in gross capital before expenses, down from the more than $620 million anticipated when the merger was announced in June. That breaks down as about $164 million from the trust account, $201 million from a PIPE led by Foxconn and $100 million in equity Agility received after the announcement. The original deal valued Agility at a $2.5 billion pre-money equity value. The company says Digit robots are already operating in nine customer environments. GXO Logistics will expand deployments to Statesville, North Carolina and Lebanon, Tennessee in early 2027, and AGCO will pilot Digit at its Hesston, Kansas plant from 2027.

Regulatory approval of a robot is one thing. Approval of a claim is another. In South Korea, a court rejected a refund suit brought by Tesla owners over the Full Self-Driving name, with 88 owners losing their bid, according to The Cool Down. Tesla has also renamed the package on European websites, Reuters reported on 29 October.

CleanTechnica, meanwhile, published a first-hand account on 1 November of a 1,500 mile trip in a 2026 Model Y running Full Self Driving Supervised V14.3.12. The account describes a full eight-stall Supercharger site in Wausau, Wisconsin and a 50 kW charge shared with an adjacent car. The author, Fritz Hasler, writes that navigation knew about a road closure but did not explain the detour in advance.

Local votes and national fleets

Google won approval for a data centre in Hermantown, Minnesota on Friday 30 October, after the city council postponed its vote twice, Tom's Hardware reported, citing MPR News. The project covers just over 200 acres and three buildings of about 1.66 million square feet. Google scaled it down and dropped an $80 million tax break request. Lawsuits and an inquiry into electricity supply remain.

Toyota and Waymo said on 30 October that they will begin testing an autonomous Toyota bZ4X fleet in Japan in 2027, CleanTechnica reported. Toyota will build the cars at its Takaoka plant, with deployment to the UK and other Waymo markets in following years. Waymo co-CEO Dmitri Dolgov said in the release that in areas where Waymo operates it has demonstrated being up to 20x safer than human drivers for serious crash injuries. Toyota executive vice president Hiroki Nakajima said safety technology "makes the biggest difference when it reaches as many people as possible".

The regulatory picture is not converging so much as multiplying: a device clearance here, a shareholder vote there, a court in Seoul, a council in Minnesota.

Elsewhere, researchers posted a benchmark on 29 October for whether self-evolving agents can learn a rule nobody told them, CLR-Bench. They found that the best gain over a frozen agent was at most 5 points, while a fixed lesson stating the convention lifted methods by 14.5 to 16.5 points. A second paper, MEMIG, argues agent memory for antibody optimisation should be indexable and quantitative rather than free-form text. A third, on auxiliary rewards, reports UID-Conj. improving an online macro-average by 2.08 percentage points over unmodified TTRL. Y Combinator-backed Momentic, meanwhile, launched on Hacker News on 2 November with an AI QA agent that claims 117,010 bugs caught across 70,607,819 test runs.