Source-linked AI summary
A Taxonomy of Construction Task Activities for Robot Workers
Sadman Sakib, Zhangyi None Peng, Yujie Pang, Yu Otsuki, Mohammad Abdullah Al Faruque
TL;DR
Construction robotics needs an explicit inventory of worker activities and executable capabilities because deployed systems remain task-specific across variable materials, tools, and sites. TARCAT builds an occupation-grounded taxonomy from O*NET tasks and instructional videos, representing work with composable primitives and parameterized skills. It defines 41 primitives and demonstrates selected elements on a DOBOT CR3 arm with a CRAFT hand, while providing a common vocabulary for demonstrations and robot capability specifications.
Problem
Construction robots require an explicit, composable specification of worker activities and capabilities because materials, tools, and sites vary substantially.
Method
TARCAT links O*NET construction tasks and instructional evidence to capability-oriented primitives, compositional skills, and a human-interpretable taxonomy.
Results
TARCAT defines 41 action primitives in 12 groups and three classes, with selected tool-use activities instantiated on a DOBOT CR3 arm and CRAFT hand.
Takeaways & Limitations
The representation provides a common vocabulary for indexing demonstrations, specifying robot requirements, organizing skill libraries, and exposing missing capabilities across embodiments.
Takeaways & Limitations
The initial hardware study is preliminary, and the taxonomy covers seven occupations with few demonstrations per task.
Abstract
from arXiv · showhide
Recent vision-language-action models offer a path toward robots with broader repertoires than conventional task-specific systems. Construction deployment, however, requires a precise inventory of worker activities and the capabilities needed to execute them. We present TARCAT, an occupation-grounded taxonomy derived from 91 O*NET tasks across seven high-employment construction occupations and 30 instructional videos of physical work. TARCAT defines 41 action primitives in 12 groups and three classes and provides a mechanism for composing parameterized primitive sequences into reusable skills. This human-interpretable structure can organize demonstrations, specify robot requirements, and support coding agents that retrieve and extend skill libraries. We also demonstrate selected primitives on a DOBOT CR3 arm with a CRAFT hand. TARCAT thereby provides a common vocabulary for analyzing human work and developing general-purpose construction robots. Annotations are available at https://github.com/AICPS/TARCAT-Taxonomy.
I. INTRODUCTION
Construction robots remain largely task-specific, creating a need for an explicit, composable specification of the capabilities required for human construction activities. TARCAT addresses this gap with reusable primitives, parameterized skills, and occupation-grounded task decompositions.
- Construction robots span varied domains, but material, tool, and site differences leave deployed systems task-specific.
- Existing construction classifications do not directly identify capability gaps or explain how reusable robot behaviors form construction tasks.
- Its skills are ordered, parameterized primitive sequences that support retrieval, adaptation, and extension by robot coding agents.
- TARCAT provides 41 action primitives in 12 groups and three classes for representing construction activities.
- The taxonomy combines demonstrations, robot requirements, and skill composition without prescribing a fixed control architecture.
II. RELATED WORK
Prior resources separately address construction operations, manipulation, recognition, or occupational scope, but they do not provide composable, robot-executable construction workflows. TARCAT bridges these levels with fine-grained primitives grounded in construction work.
- Existing construction studies demonstrate individual capabilities but do not define the action space of a generalist construction robot.
- General video labels omit trade context and ordered workflows, while motion datasets omit robot-executable tool and contact skills.
- TARCAT combines fine-grained primitives with compositional skills grounded in construction work.
- Its occupational grounding connects the taxonomy to economically significant occupations rather than a manually selected set of robot demonstrations.
III. INSTRUCTIONAL VIDEO COLLECTION
TARCAT grounds its construction corpus in U.S. Department of Labor occupational resources and instructional demonstrations. The resulting videos cover 91 O*NET tasks across seven selected construction occupations.
- The corpus uses O*NET occupational task statements and BLS 2024 employment data to select seven high-employment construction profiles.
- 169 O*NET task statements comprise the selected occupations, including 137 physical-movement tasks and 32 non-movement tasks.
- 30 instructional videos cover 91 O*NET tasks across the selected occupations.
- Videos were retained when they showed clearly observable, task-relevant actions and were mapped to every demonstrated task.
IV. ACTIVITY TAXONOMY
TARCAT abstracts construction work into capability-oriented action primitives and composes them into reusable, context-independent skills. This structure preserves control-relevant distinctions while supporting transfer across trades.
- Action primitives are context-independent actions requiring distinct robot capabilities and composable into complex tasks.
- Reusable skills are ordered primitive sequences that achieve a subgoal across tasks and occupations.
- Primitives are distinguished by exercised capability rather than occupation or tool identity.
- Different tools share a primitive when they impose the same motion or interaction requirement, while differing contact, force, or coordination patterns remain distinct.
A. Primitives
TARCAT organizes construction activities into 41 primitives across three broad classes, with groups collecting shared functions or motion patterns. The representation links occupational activities to observable execution and reusable capabilities.
- A. Primitives: 41 primitives are organized into information processing, communication, and tool use classes.Classes distinguish broad roles, while groups collect primitives with shared function or motion patterns.
- A. Primitives: The taxonomy uses object for a tool or manipulated item, while tool denotes cases without a separate object, enabling generalization across trades and equipment.
- A. Primitives: Information processing covers assessing conditions, interpreting information, and planning resources, tools, designs, or layouts.
- A. Primitives: Communication covers coordinating activities with co-workers and creating written documents that preserve task-relevant information.
- A. Primitives: Tool use includes finger manipulation, arm positioning, wrist rotation, mixing, surface contact, locomotion, and whole-body load handling.
B. Skill and Task Decomposition
TARCAT represents reusable skills as ordered primitive sequences and composes skills and standalone primitives into construction procedures. Variables instantiate context-independent skills, while procedure representations capture repetition, simultaneity, and nesting.
- B. Skill and Task Decomposition: A skill is an ordered sequence of activities labeled with action primitives and may contain variables instantiated with observed tools or objects.The wire-disconnection example uses a screwdriver and terminal as task-specific variables.
- B. Skill and Task Decomposition: The wire-disconnection skill decomposes loosening a terminal screw and pulling the wire into wrist-rotation and arm-motion primitives.
- B. Skill and Task Decomposition: Construction procedures compose skills and standalone primitives while explicitly representing simultaneous actions, repeated actions, and nested subskills.Simultaneous actions are recorded sequentially, repetition uses a repeated attribute, and skills may invoke subskills.
V. REAL-WORLD DEPLOYMENT
The authors instantiated four tool-use activities on a DOBOT CR3 arm with a CRAFT hand. The demonstrations covered positioning, placing, hammering, and brushing behaviors represented by selected TARCAT primitives.
- V. REAL-WORLD DEPLOYMENT: Four tool-use activities were instantiated on a six-axis DOBOT CR3 arm with a 15-actuator, tendon-driven CRAFT hand.
- V. REAL-WORLD DEPLOYMENT: The scripted demonstrations included positioning a power screwdriver, placing a hammer in a box, hammering a nail, and brushing objects aside.
- V. REAL-WORLD DEPLOYMENT: These demonstrations spanned arm translation and positioning, arm and wrist rotation, and surface-contact manipulation.
VI. CONCLUSION
TARCAT provides an occupation-grounded, human-interpretable specification of construction capabilities through primitives and parameterized skills. Its decompositions can support instruction-to-skill mapping and reuse, while the current validation remains limited in scope.
- VI. CONCLUSION: TARCAT links O*NET tasks with instructional evidence to represent construction work through 41 primitives and parameterized skill composition.
- VI. CONCLUSION: The taxonomy translates occupational work into a structured capability specification for analyzing demonstrations and robot requirements.
- VI. CONCLUSION: Its decompositions can guide agents in mapping instructions to primitives, composing task-specific skills, and retaining validated compositions for reuse.
- VI. CONCLUSION: The hardware study is preliminary, with coverage limited to seven occupations and few demonstrations per task.Future work will expand the corpus, assess annotation agreement and coverage, implement more primitives on additional embodiments, and evaluate long-horizon composition.