The publication counted 1,933 papers on the 2026 program, versus 2,672 a year earlier. It reports 4,348 submissions and 1,585 accepted papers, for a 36.5% acceptance rate, down from 46% in 2025.
The analysis says papers with at least one US-based author rose from 559 to 727, while papers with mainland Chinese authors fell from 1,469 to 640. Tsinghua University remained the institution appearing on the most papers, with 88; Carnegie Mellon University had 74.
Robot learning appeared in 37% of the 2026 papers, up from 31%. Paper titles mentioning vision-language-action rose from four to 65; titles naming LLMs, VLMs, or language-guided control rose from 4.4% to 7.3% of the program.
The source does not establish why acceptance, country representation, or topic shares changed. It notes that host location and travel costs are plausible factors, and that title matching is only a lower-bound measure of research use.