**Language**: English | [한국어](TAXONOMY_EVALUATION.md) # Taxonomy Self-Evaluation: vs. Springer Handbook of Robotics > This document evaluates how MECE (Mutually Exclusive, Collectively Exhaustive) and insightful our 4-level phylogenetic taxonomy (Phylum > Class > Order > Genus) is, compared to the de facto standard reference in robotics — the **Springer Handbook of Robotics** (2008 and 2016 editions). --- ## Springer Handbook of Robotics — Structure Overview The Handbook, edited by Siciliano & Khatib, is the largest comprehensive reference in robotics and maintains a 7-part structure across both editions. ### 2008 Edition (64 chapters, 7 parts) ``` Part A: Robotics Foundations — Kinematics, Dynamics, Mechanisms, Sensing, Motion Planning, Control Part B: Robot Structures — Manipulators, Parallel, Legged, Wheeled, Aerial, Underwater Part C: Sensing and Perception — Vision, Force/Tactile, Sonar, GPS, SLAM Part D: Manipulation and Interfaces — Grasping, Assembly, Haptics, Teleoperation, HRI Part E: Mobile & Distributed — Navigation, Multi-robot, Swarm, Network Part F: Field & Service Robotics — Medical, Mining, Construction, Agri, Space, Underwater Part G: Human-Centered & Life-like — Humanoids, Social, Cognitive, Bio-inspired, Microbotics ``` ### 2016 Edition (83 chapters, same 7 parts + expansion) Same structure with added chapters: Cloud Robotics, Autonomous Vehicles, Surgical Robotics, etc. --- ## MECE Comparison ### Mutual Exclusivity (ME) Violations in the Handbook | Topic | Overlapping Parts | Problem | |---|---|---| | Aerial Robots | Part B (Robot Structures) **&** Part E (Mobile) | Flying robots are both a "structure" and a "mobile system" | | Underwater Robots | Part B **&** Part F (Field) | Underwater robots appear in both structural and service chapters | | SLAM | Part C (Sensing) **&** Part E (Mobile) | Localization spans both perception and mobile robotics | | Medical Robotics | Part D (Manipulation) **&** Part F (Service) | Surgical robots are both manipulation technology and service applications | | Swarm Robotics | Part E (Distributed) **&** Part G (Bio-inspired) | Swarm behavior is classifiable under both distributed systems and bio-inspiration | **Root cause:** The Handbook's organizing principle is a mixture of `technology layer (Part A)` × `robot type (Part B)` × `functional domain (Parts C–E)` × `application domain (Parts F–G)`. A single paper can simultaneously belong to multiple parts — a structural MECE failure. ### Collective Exhaustiveness (CE) Gaps | Missing / Underrepresented Area | Notes | |---|---| | **Learning for Robotics** | No independent part in 2008. RL and deep learning are scattered after Part A | | **Theoretical Foundations** | Kinematics, optimization, Lie groups are bundled into Part A "Foundations," obscuring the mathematical lineage | | **Foundation Models** | Even the 2016 edition cannot accommodate VLA, World Models, Diffusion Policy (2023+ trends) | | **SLAM as independent domain** | Despite being a distinct research community, SLAM is subordinated to Sensing | --- ## Strengths of Our Taxonomy ### 1. Consistent Organizing Principle Our taxonomy defines every Phylum by a single principle: **"what capability does this research enable?" (functional capability)**. ``` Phylum Principle ───────────────────────────────────────────────────── Perception & Sensing → Detecting and recognizing the environment SLAM & Localization → Self-localization and map building Planning → Action plan generation Control → Real-time actuation control Manipulation → Object manipulation Locomotion → Platform movement Robot Design & Hardware → Physical system design Human-Robot Interaction → Human-robot interface Multi-Robot Systems → Multi-agent coordination Learning for Robotics → Learning-based capability acquisition Application Domains → Application contexts (cross-cutting) Theoretical Foundations → Mathematical and theoretical basis Robot Software & Arch. → Software architecture ``` ### 2. Key Design Decisions That Work Better **Separating SLAM from Perception** - The research communities are genuinely distinct (ICRA SLAM sessions ≠ vision sessions), as are the methods. - The Handbook subordinates SLAM to Sensing, erasing this distinction. **Elevating Learning for Robotics to a top-level Phylum** - Our classified data shows ≈354 papers (4.7%) in this Phylum across all years; this proportion grows sharply post-2020. - The 2008 Handbook essentially ignored this domain; the 2016 edition still gave it no dedicated part. **Explicit cross-cutting treatment of Application Domains** - Autonomous Driving, Medical Robotics, and Space Robotics are *application contexts*, not technologies. - Isolating them as a separate Phylum prevents contamination of the technology-based classification (Perception, Planning, …). **Decoupling Theoretical Foundations** - Kinematics, dynamics, Lie groups, stability theory, and formal verification no longer drown inside Control or Planning chapters. - The lineage of pure-theory research can now be traced independently. ### 3. Reflecting 2020s Research Trends | Trend | Our Classification | In the Handbook | |---|---|---| | VLA / π₀ / RT-2 | `Learning > Foundation Models > VLA` | Not present | | Diffusion Policy | `Learning > Foundation Models > Diffusion Policies` | Not present | | Gaussian Splatting SLAM | `SLAM > Neural Implicit / GS SLAM` | Not present | | Event Camera | `Perception > Event-based Vision` | Not present | | Safety-Critical Control (CBF) | `Control > Safety-Critical Control` | Not present | | Loco-Manipulation | `Manipulation > Mobile Manipulation` | Not present | --- ## Where the Handbook Wins (Honest Assessment) | Area | Why the Handbook Excels | |---|---| | **Robot-type intuitiveness** | Chapter-per-robot-type (Humanoid, Quadruped, Aerial) is natural for browsing by embodiment | | **Application-first search** | Non-specialists looking for "Medical Robotics" find it immediately | | **Historical narrative** | Each chapter can tell the story of a subfield's development — good for a textbook | | **Authority** | Written by 164 contributors, peer-reviewed, industry-standard reference | However, these strengths follow from the Handbook being designed as a **textbook (pedagogy)**, not as a paper classification system. The structural flaws listed above are critical when the goal is **automated, MECE classification of 7,000+ papers**. --- ## Verdict Our 4-level phylogenetic taxonomy outperforms the Springer Handbook of Robotics structure on the following dimensions: 1. **MECE** — A single, consistent organizing principle (functional capability) minimizes mutual exclusivity violations. 2. **Insightfulness** — The SLAM/Perception split, Learning as an independent Phylum, and explicit Theoretical Foundations reflect the 2020s research landscape more accurately. 3. **Currency** — Foundation Models, Gaussian Splatting, Event Cameras, and other post-2023 trends are accommodated in the taxonomy. The most significant differentiator is **elevating Learning for Robotics to a first-class Phylum**. Both Handbook editions failed to give this domain an independent standing, whereas our taxonomy correctly captures where the field's center of gravity now sits. The Handbook is undeniably more intuitive in some respects, but that is a consequence of its pedagogical design goals. For the purpose of **MECE classification of 5,000+ papers**, our taxonomy is the better fit. --- *Reference: Springer Handbook of Robotics, 2nd ed. (Siciliano & Khatib, 2016), ISBN 978-3-319-32550-7; 1st ed. (2008), ISBN 978-3-540-23957-4.* *ToC PDFs used for the comparison (cover + table of contents only, for fair-use review):* - [`references/springer_handbook_robotics_2008_toc.pdf`](references/springer_handbook_robotics_2008_toc.pdf) — 1st edition (2008) - [`references/springer_handbook_robotics_2016_toc.pdf`](references/springer_handbook_robotics_2016_toc.pdf) — 2nd edition (2016)