Robotics Research Trends

📐 How this analysis works

A data-driven map of what robotics & control research is actually working on — built from arXiv papers (cs.RO Robotics + eess.SY Systems & Control) spanning . An automated pipeline collects, reads, categorizes, and de-duplicates the literature, then tracks how topics rise and fall over time.

1. Collect. Papers are pulled from arXiv (Robotics + Systems & Control) for a curated set of leading robot-learning groups.
2. Classify. A large language model reads each paper's title and abstract and tags it across five taxonomies — research domains, algorithms, network architectures, simulators/frameworks, and hardware/morphologies.
3. Mine the full text. To catch tooling an abstract leaves out, each paper's arXiv LaTeX source is downloaded and scanned for known simulators and frameworks.
4. Add affiliations. Each paper is matched against OpenAlex to label its authors as academia or industry — this powers the academia-vs-industry comparison.
5. Consolidate the vocabulary. A second LLM pass folds synonyms and spelling variants into a single canonical term per concept (e.g. "RL" → "Reinforcement Learning"), so one trend isn't split across aliases.
6. Visualize. The cleaned data drives the interactive views below — how topics co-occur, which are heating up or cooling down, a lifecycle map, an academia-vs-industry split, and a year-by-year explorer.

Coverage reflects a curated set of leading robot-learning groups — a focused look at the field's frontier, not an unbiased sample of all robotics research.