Short answer
Predetermined motion time systems such as MTM break manual work into basic motions like reach, grasp, move, position and release, each with a standard time. Industrial engineering teams hold these breakdowns for assembly and packing tasks. They give a human-level vocabulary of action primitives with timing, but they describe a trained operator's hands, not robot actions, and need linking to video to be useful.
The buyer's problem
Robot-learning teams need labels for manipulation segments: where does a "reach" end, where does a "grasp" begin, how long does each take. Annotators can produce these labels from video, but the result depends on how carefully the vocabulary was defined. Industrial engineers have used a standardized vocabulary for decades to plan and time manual work.
MTM, Methods-Time Measurement, is one widely used example. Its association describes it as a method standard for describing, evaluating and designing human work, and training materials describe it as analyzing manual operations into basic motions, among them reach, grasp, move, position and release, with predetermined time standards (MTM Association). The association was founded in 1962 and reports more than 240 corporate members, so analyses may sit with many manufacturers and consultancies. MOST and other predetermined-time systems follow related logic at coarser levels of detail. This page is about locating and judging those records. For goal-level decomposition instead of motion-level, see hierarchical task analyses.
What a motion-time record contains
A work-measurement study produces a row per motion element with:
- Element code, drawn from the system's tables, which encode the motion type and variables such as distance, case of grasp or accuracy of placement.
- Sequence order within a work cycle, and often a left-hand and right-hand column.
- Time value, in system units. For MTM the unit is the time measurement unit: 1 TMU equals 0.00001 hour, or 0.036 seconds (MTM Association).
- Totals for the cycle and an allowance for fatigue or delay, set by the employer.
- Context: workstation layout, part and tool names, and the method description.
The codes are system-specific shorthand, not natural language. A short reach-and-grasp sequence might read as a few alphanumeric codes with distance classes; the analyst's sheet usually adds a text description. Records may be in spreadsheets, specialist software exports or printed worksheets.
Why they map onto action primitives
The motion vocabulary is close to what manipulation researchers label. A reach, a grasp and a move line up with common segmentation in video datasets, such as the goal and step annotations in Ego4D Goal-Step, which adds hierarchical goal and procedural-step annotations to Ego4D with 48K procedural step segments (Song et al., 2023). The two are at different levels: Goal-Step labels meaningful task steps, while motion-time elements are finer, sub-step units. A motion-time record can supply a controlled, expert-defined vocabulary for the finer level, and times that reveal which steps dominate a cycle.
What this data can support, and what it cannot
It can support:
- Label vocabularies. A defined set of primitives with decision rules for variants.
- Segment priors. Typical durations and orderings help check whether automatically segmented video looks plausible.
- Task decomposition below step level. Elements show how a step such as "insert connector" splits into reach, grasp, move, position and release.
- Workstation context. Distances and layouts from the study hint at reach geometry.
- Human baselines. A standard time is a reference for comparing robot cycle times, though not a target a robot should assume it can meet.
It cannot support:
- Robot actions. The codes are about a human's hands. They contain no joint states, commanded motion or forces. Moving from human motion to humanoid control still runs through the embodiment gap.
- Contact and force. "Apply pressure" carries no magnitude.
- Perception. Nothing shows what the operator saw.
- Observed variation. Standard times describe a trained operator following a defined method at normal pace, not the spread of real operators or the errors that occur.
- Anything without the underlying method. A sheet of codes without the method description and part names is almost opaque outside the originating team.
Illustrative record with video linkage
# Illustrative - synthetic codes and values
study_id: wm_0027
task: "Insert connector C and secure latch"
system: MTM-1
cycle:
- seq: 1
hand: right
element_code: "R-B-12" # synthetic code
description: "reach to connector bin, 12 in"
time_tmu: 12.9
video_ref: {clip: clip_0412, t_start: 3.20, t_end: 3.70}
- seq: 2
hand: right
element_code: "G-1A"
description: "grasp connector"
time_tmu: 2.0
video_ref: {clip: clip_0412, t_start: 3.70, t_end: 3.78}
method_notes: "fixture_F1, bin at left of work area"
analyst: analyst_09
date: "synthetic"
The video_ref block is the field that turns a work-measurement sheet into a labeling aid. Most existing files lack it, and creating it requires someone to align codes to footage. See SOP-to-video step alignment for the alignment method.
Where to find them
| Holder | Typical context | Note |
|---|---|---|
| Manufacturers' industrial engineering departments | Line balancing, standard-time setting | Often tied to labor agreements, so access is sensitive |
| Industrial engineering consultancies | Client time studies | Consultancy may own the analysis |
| Distribution and packing operations | Engineered labor standards | Coarser, task-level records |
| Training organizations | Teaching examples | Generic, not site-specific |
The manufacturing data hub covers the assembly side of this supply. Workplace time studies also often sit beside video taken for the study itself, which raises consent questions covered below.
Questions to ask suppliers
- Which system (MTM-1, a higher-level variant, MOST, or in-house) produced the records, and who certified the analysts?
- Are element codes accompanied by text descriptions and method notes?
- Were studies done from video, live observation or both, and can any video be licensed?
- Do records have left-hand and right-hand columns, and are both populated?
- How many distinct tasks, operators and workstations are covered?
- Are standard times used in labor agreements, and does that restrict disclosure?
- Who owns the studies: the employer, a consultancy or a software vendor?
- Which fields identify individual operators, and have they been removed?
- How old are the studies, and have methods changed since?
Permissions and provenance
Time studies relate to people's work pace. Employers may treat them as sensitive labor data, and operators may have been filmed for the study without consent for model training. Consultancy contracts may reserve ownership. This is not legal advice. Ask whether the supplier can license the records, any linked video, and the derived labels, and see worker consent and identifiable people. Removing names does not settle consent for the footage.
Limitations and additional preparation
- Granularity mismatch. Motion elements are finer than most task-step labels; buyers should decide whether they need that level before paying for alignment.
- Code decoding. Specialist knowledge is needed to translate codes to usable labels.
- Small, scattered holdings. Expect tens of studies at a plant rather than thousands.
- Standard time is not observed time. Do not treat it as ground truth for duration labels.
- Alignment cost. Matching codes to video frames is manual work and should be budgeted, with a sample audit.
- Layered use. Pair with step-level labels and, for control, with robot data; this asset supplies vocabulary and timing, not actions. Contrast it with occupational task taxonomies and step granularity choices.
Reading a time study as a labeling guide
When a study is available with its video, an annotator can use the element sequence as a template: the sheet says the cycle has a reach, a grasp, a move, a position and a release, so the annotator looks for those five segments and marks boundaries. This constrains the labeling task and speeds review, because a missing or extra segment is easy to spot. Disagreements between the sheet and the footage deserve a note rather than a silent fix, since they show where the operator departed from the documented method.
A caution on timing: standard times describe normal pace for a trained operator, so use them to sanity-check segment durations, not to set them. A segment that runs several times longer than the standard may indicate a hesitation, a regrasp or a mislabeled boundary, and each is worth looking at before it enters a training set.
Discuss your requirements
If you want motion-element breakdowns for specific manual tasks, describe the task types, level of detail and permitted uses to SourceX. SourceX looks for US businesses that hold the material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX follows up about potential sources. Start at the humanoid training data guide or the procedures hub. The catalog page on manufacturing quality records describes nearby material.
Explore sourcing and licensing options with SourceX.
Sources
- MTM ASSOCIATION e. V. (Methods-Time Measurement) — MTM as a method standard, basic motions, TMU definition, membership.
- Ego4D Goal-Step: Toward Hierarchical Understanding of Procedural Activities — 48K procedural step segments in hierarchical annotation.