A trainable sample aligns at least four things: time, observation, action, and body identifier. The two-finger parallel gripper RMG22-65 provides the body and part of the action carrier among these. Communication interface: 6-pin aviation connector. Communication protocol: RS485 / Modbus-RTU. Operating environment: 0–50 °C, relative humidity ≤85%. There is no camera on the page. Which channel the observation comes from is governed by this sentence. A video without timestamps is only a recording.
Data Recording
The action field is built separately from degrees of freedom, chassis, and end effector. This is an end-effector gripper, not a multi-joint arm. The public table does not give the gripper a single "degrees of freedom" integer; do not invent one to align with other embodiments. Open / close time 2 s. Position repeatability ±0.05 mm. The 2 seconds is the time basis for opening or closing, not the control frequency. It is itself the end effector. Whether it directly matches the body's flange, wiring harness, and protocol must be confirmed in the configuration sheet; the page states that plug-and-play is not assumed. Quantities not present in these three sentences must not be written into the data dictionary as already collected. If gripper opening and closing uses RMG22-65, the opening and closing follows the gripper page's 2-second basis, not one dimension of the body's degrees of freedom, unless the recording scheme explicitly encodes it as one dimension and clearly states the unit.
A human's head-mounted device and gloves can operate in parallel, but cannot be labeled with the motion labels of this machine body. Cross-embodiment transfer is a research topic, not a feature already completed on this page. Works such as Open X-Embodiment require the embodiment to be explicitly labeled, which precisely shows that sensors from different pages cannot be merged into one robot.
The quality statement reads: the action space of this sample equals the written configuration for this order, not the union of all options on the two-finger parallel gripper RMG22-65 product page. If an option was not purchased, it is not in the sample.

How to Use This Page in This Version
The data dictionary starts with four columns: time, observation, action, and body. The body column records the two-finger parallel gripper RMG22-65 and the configuration sheet number for this session; do not write "teleoperation platform." The number of channels in the observation column comes from: communication interface 6-pin aviation connector. Communication protocol RS485 / Modbus-RTU. Operating environment 0–50 ℃, relative humidity ≤85%. There is no camera on the page. The joints in the action column come from: this is an end-effector gripper, not a multi-joint arm. The public table does not give the gripper an integer "degrees of freedom" value; do not invent one to align with other embodiments. The chassis comes from: open / close time 2 s. Position repeatability ±0.05 mm. The 2 seconds is the time for opening or closing, not the control frequency. The end-effector comes from: it is the end-effector itself. Whether it directly matches the body's flange, wiring harness, and protocol must be confirmed in the configuration sheet; the page states that it is not plug-and-play by default. For any of the four columns that are not on the page, the dictionary writes "not collected," not a default value.
File naming should include date, robot unit, and configuration sheet number. Do not use non-traceable terms like "final version" or "new data." If the end effector or kit column changes on the same day, the configuration sheet number must change; otherwise, samples from two different action spaces will end up in the same folder.
Human-side devices go in a separate directory. Head-mounted video, glove posture, and robot joint data must not be merged into a single training file when they are not aligned. Merging is a subsequent step, contingent on the synchronization scheme having been specified. This page does not promise merging as a factory function.
The quality note written for humans repeats the boundaries: if the optional camera was not purchased, it is not in the observation column; if the recommended gripper is not specified, it is not in the action column; the success rate in the paper is not a label for this batch of data. /product-gripper-2f.html is the basis for the body column.
Line-by-Line Verification
Degrees of freedom. This is an end-effector gripper, not a multi-joint arm. The public table does not give the gripper an integer "degrees of freedom" value; do not invent one to align with other embodiments. This row only states the number and locations of joints that have been made public. It does not evaluate intelligence, and it does not automatically include the waist, legs, neck, or whole-machine total that the page does not name. When writing the action vector, only build channels for the parts listed in this row; parts not listed remain blank. Do not write blank as 0, because 0 will be treated by the training script as a real angle. Do not fill in the degrees-of-freedom integer from another device after this row to make it add up. If the master–slave joint order is not given on the page, mark it as pending written configuration confirmation; do not number it yourself. The joint set in the action column equals this row, no more and no less. The action vector builds channels only for the parts named in this row; parts not named are left blank, not filled with 0.
Load and mass. Maximum envelope grasp load 5 kg. Recommended maximum friction load 2 kg. Maximum gripping force 130 N. The three numbers cannot all be written as the same 5. The numbers in this row must be read with their subject: rated, self-weight, whole-machine mass, envelope, friction, or gripping force. If the subjects differ, they cannot be merged into one number, nor can conversions not given on the page be performed. If the object in the field is heavier, the specification cannot be changed retroactively; a larger specification also does not prove that every grasping method reaches that limit. The tool mass is undetermined; do not write the arm's rated value as the graspable object mass. The acceptance table retains the original units. If the mass label uses the rated value, state that it is the specification upper limit, not the object weight of every sample. During acceptance, read the subject first and then the number; if the subjects do not match, split into two rows.
Form factor and scale. Overall dimensions 112 × 60 × 167 mm. Stroke 65 mm. Finger height 80 mm. Platform weight 745 g. This row describes form factor or structural scale, used to verify doors, elevators, tabletops, or transport — it is not operating height, nor is it arm reach. Do not add it to the vertical range. If the diagram shows protrusions, antennas, or human standing positions not present in this row, note separately that they are not part of the specification. If weight appears in an adjacent row, do not write it into the dimension sentence. Form factor does not enter the action vector. Doors, elevators, and transport are verified only against form factor; do not treat this row as operating height.
Operating range. The stroke of 65 mm is the finger opening and closing distance, not the robotic arm's reach, and not the body's vertical working range. This row describes the range the hand or lift can reach; the definition must follow the original wording: vertical working range, reach, motion radius, or stroke — these cannot be added together or renamed. If "excluding end-effector" is written, it must remain in the figure caption. When the original text does not mark a unit, retain the note "unit not specified"; do not delete it for tidiness. The working range can be written into task metadata, not into joint angles. The figure caption retains the range definition from the original wording, as well as the phrases "excluding end-effector" or "unit not specified."
Motion. Open/close time 2 s. Positional repeatability ±0.05 mm. The 2-second figure is the open-or-close duration, not the control frequency. This line describes how the ground platform or primary arm moves. Four-wheel four-drive, differential drive, in-place turning radius, and rotational speed are terms for different mechanisms and are not interchangeable. Published speed is a capability, not a recommended teaching speed; teaching may be slower, but the slower value is recorded under experimental conditions and is not written back into the specification. Pages that do not state gradeability or positioning must not borrow from other pages. Whether a chassis passage exists is determined by whether this line describes locomotion. Teaching speed may be lower, but the lower value is recorded under operating conditions and does not change the published capability in this line.
Power supply. Rated voltage 24 V. Rated / maximum current 1.9 / 5.7 A. Rated / maximum power 45.6 / 136.8 W. There is no battery capacity and no runtime in hours. This row only retains the electrical terms that appear on the page: voltage, capacity, runtime, charging, current, or power. If there are no "hours," write that runtime is not given; do not fill in hours from an adjacent page. Do not rename charging time as runtime. Rated and maximum must be in separate columns. Retain "approximately" for approximate values. Do not sum the ampere-hours of multiple devices and call it one device's battery. The battery row does not produce sensor timestamps. Hours, charging, or current that do not appear remain not given; do not copy them from an adjacent page.
Perception and computation. Communication interface 6-pin aviation connector. Communication protocol RS485 / Modbus-RTU. Operating environment 0–50 ℃, relative humidity ≤85%. There is no camera on the page. This row takes inventory of the items already specified in sensing, computation, or communication, and retains the words standard, included, optional, and selectable. An optional item that is not checked does not count. When the cover and the list conflict, cite the list, while not pretending the cover statement does not exist. Head-mounted devices and gloves do not enter this row. Processors, hard-drive capacity, or camera channel counts that are not written out must not be drawn into the system diagram. The topics in the observation column are consistent with the channel counts in this row. Optional sensors that were not purchased do not enter this data dictionary.
End effector. It is itself the end effector. Whether it directly matches the flange, wiring harness, and protocol of the robotic platform must be confirmed in the configuration sheet; the page states that it is not plug-and-play by default. This row describes where contact or grasping occurs. Quick-change, compatible, and recommended are not models already included in the box. The ring fixture on the main arm is not an end-effector gripper. Force, stroke, and payload remain as written in their respective rows and are not merged into a single system metric because they are mounted on a particular robotic platform. For flange and protocol mappings not provided, write "to be confirmed," not "plug-and-play." The end-effector status only enters the action column once the model is determined. "Supported" or "compatible" without a model cannot be treated as delivered.
Boundary. The standard quantity table for the lift model does not include this gripper. The folding model's description is about quick-change capability, not that RMG22-65 is already in the box. The recommended item CTAG2F90-D is also not this model. This line is a boundary deliberately separated on the page: another model, another column, another product page, or an interface that still requires project confirmation after modularization. The boundary is not a defect note; it is to prevent verbally buying two pages as one. When writing a proposal, copy the boundary into "not included in this order". If a device not included is later added, open a new configuration order; do not go back and change the platform name in old samples. Devices not included should not appear in the platform identifier. Devices within the boundary should be ordered separately; do not rename old samples to say they already include it.
Configuration column. Enter the model number RMG22-65 in the order, and specify which robotic platform it is installed on, the quantity, and whether it replaces other gripper models listed in that platform's recommended column. This row distinguishes standard, recommended, and optional. Recommended and optional items are not delivered unless checked. When the standard row and the parameter table row are inconsistent, keep both rows and use "locked for this order" to indicate which one is actually delivered, rather than striking out one row in the order without signing off. Price and lead time are not written into this row's restatement. The sample's configuration sheet number must explain whether the recommended item is checked. Unchecked recommended and optional items are written as not delivered in the order.
Excerptable. Copy the attachment line by line: envelope 5 kg; friction recommendation 2 kg; maximum gripping force 130 N; stroke 65 mm; finger height 80 mm; 745 g; open/close 2 s; repeatability ±0.05 mm; 24 V; 1.9/5.7 A; 45.6/136.8 W; RS485 / Modbus-RTU. This is the type of line that can be pasted into the attachment: it has a subject, has units, and can be traced back to the original sentence on the product page. When pasting, keep words such as approximately, excludes, not specified, optional, and recommended. Do not convert and keep only the result. If the same number has different meanings on two pages, write it as two separate lines in the attachment. Do not mix test protocols into the specification excerpt. The data description should quote the original sentence from the attachment and not write a separate set of numbers. When pasting into the attachment, paste words like approximately, excludes, and optional along with them.
Do not write. Do not convert 130 N to 5 kg and delete the original row. Do not write ±0.05 mm as the repeatability of the attached robotic arm. Do not write runtime. This row is a negative statement for review purposes. It does not describe what the device can do; it only prevents hours, degrees of freedom, load subjects, or cover slogans from adjacent pages from being written in. If a negative statement is hit, revert to the original product page wording; do not interpret it as approximately correct. External paper success rates and total hours not published on this site likewise must not be written in. If a number from a negative statement appears in the dictionary, judge it as mixed writing. If a manuscript search hits a negative statement, revert to the original product page wording.
| Projects | Public-Facing Positioning of This Page |
|---|---|
| Degrees of Freedom | This is an end-effector gripper, not a multi-joint arm. The public table does not give the gripper a "degree of freedom" integer; do not fabricate one to align with other platforms. |
| Load and Mass | Maximum envelope grasp payload 5 kg. Recommended maximum friction payload 2 kg. Maximum clamping force 130 N. These three numbers cannot all be written as the same 5. |
| Form Factor and Scale | Overall dimensions 112 × 60 × 167 mm. Stroke 65 mm. Finger height 80 mm. Body weight 745 g. |
| Operating Range | The 65 mm stroke is the finger opening distance, not the robotic arm reach, nor the platform's vertical working range. |
| Motion | Open / close time 2 s. Positional repeatability ±0.05 mm. The 2 seconds is the specification for the open or close time, not the control frequency. |
| Power | Rated voltage 24 V. Rated / maximum current 1.9 / 5.7 A. Rated / maximum power 45.6 / 136.8 W. No battery capacity, no battery life in hours. |
| Perception & Compute | Communication interface: 6-pin aviation connector. Communication protocol: RS485 / Modbus-RTU. Operating environment: 0–50 °C, relative humidity ≤85%. There is no camera on the page. |
| End Effector | It is itself the end-effector. Whether it directly matches the flange, wiring harness, and protocols of the platform must be confirmed on the configuration sheet; the page states that plug-and-play is not the default. |
| Boundary | The standard quantity table for the lift model does not include this gripper. The folding model lists quick-change capability, not that the RMG22-65 is already in the box. The recommended item CTAG2F90-D is also not this model. |
| Configuration Panel | The order should specify model RMG22-65, and state which platform it is mounted on, the quantity, and whether it replaces other gripper models listed in that platform's recommended column. |
| Excerptable | Attachment specifications listed line by line: envelope 5 kg; recommended friction 2 kg; maximum gripping force 130 N; stroke 65 mm; finger height 80 mm; 745 g; open/close time 2 s; repeatability ±0.05 mm; 24 V; 1.9/5.7 A; 45.6/136.8 W; RS485 / Modbus-RTU. |
| Do not write | Do not convert 130 N to 5 kg and then delete the original row. Do not write ±0.05 mm as the repeatability of the attached robotic arm. Do not state battery life. |

Relationship to Public Methods and Adjacent Pages
The embodiment is an actuator, not a dataset. Running the machine once does not automatically produce trainable action labels. A trainable demonstration must at least be alignable on: time, observation (which camera stream, resolution, frame rate), action (which of the joints, gripper, chassis), and the identifier of that specific physical unit. If any one is missing, the file is merely a log, not an imitation learning sample.
A person wearing E6 or E2 walking around produces the human's head pose and hand data, not the motion of a robot end effector. The public documentation for E2 does not include on-board SLAM; poses are reconstructed offline. A 4 Hz preview is not training footage. E6 training should read the 60 Hz raw stream, not the 10 Hz preview. The G7 and G2-L gloves record the hand, not the joints of this rig. These devices can be used simultaneously, but the files serve different roles.
The "success rate" in strategy papers is computed on the authors' own tasks and data. Writing ALOHA or Diffusion Policy results as "achievable with this embodiment" mistakes method literature for an acceptance report.
This article lists: the fields that must appear on the recording sheet if this embodiment is used for demonstration, and which adjacent products' data cannot impersonate its motions.
Before handing off to a colleague
During data dictionary review, ask four questions for each column: where does the time come from, how many observation streams are there, what action channels exist, and which configuration sheet does the rig belong to. If any column's answer is "approximately," the dictionary is returned. Human-side directories and rig directories are stored separately. When the configuration sheet number changes, new samples must not be written into the old directory. The last sentence of the quality note is repeated: optional items not purchased are not in the samples, and the paper's success rate is not a label. The dictionary version is submitted together with the product page verification date. In directory permissions, unsynchronized and signed-configuration data are placed under two separate root directories. The training account has read-only access to signed configurations by default. The root directory name for human-side data contains "human"; the rig directory name contains the rig and configuration sheet number. Directories without numbering are cleaned up weekly and not transferred into the training set. The required fields in the quality note template are: time, number of observation streams, action channels, and rig number — the form cannot be saved if these are left blank. The save button does not offer "fill in later."
First, if the four columns cannot be answered, return the dictionary.
Second, the human side and the robot side get separate root directories.
Third, whenever the configuration number changes, switch directories.
Fourth, directories without numbering do not enter the training set.
Fifth, required fields cannot be left blank when saving.
Sixth, a paper's success rate is not a label.
Continue Reading
- 01 Task Boundaries
- 02 Specification Standards
- 03 Degrees of Freedom
- 04 Payload & Self-Weight
- 05 Workspace
- 06 Locomotion Method
- 07 Perception and Computing
- 08 Power Specifications
- 09 Data Recording
- 10 Time Alignment
- 11 End-Effector Boundaries
- 12 Site Constraints
- 13 Configuration Sheet
- 14 Three-Way Comparison
- 15 Literature References
- 16 Master-Slave Workstation
- 17 Acceptable Deliverable Statements
- 18 Prohibited Conflations
- 19 On-Site Deployment
- 20 Evaluation Path
Back to Product Center View this product page
This article was written on October 3, 2026, for selection and specification verification purposes. Device parameters are subject to the product page tables and do not constitute a supply commitment. External papers only illustrate how problems are framed in the industry; their success rates and data scales belong to the original authors' experiments and are not TOBI's measured results. Phone: +86 137 5502 0164.