The task must first be written as three verifiable sentences: where the operating height falls, whether the object mass is covered by the rated value, and whether the chassis moves during teaching. The public sentences available for checking a wheeled humanoid teleoperation main body are: vertical arm reach 2100 mm. Horizontal arm reach 1818 mm, excluding end effector. Single-arm payload 5 kg. Whole-machine weight approximately 125 kg, not single-arm payload. Movement speed 0.1–1.5 m/s. Turning radius 690 mm, specified as in-place rotation. The three sentences correspond to height, mass, and ground respectively. If any one of them does not cover the on-site conditions, do not write this device as the default platform for that task on the proposal cover.
Task scope
Where the motion comes from is a separate task statement. The wheeled humanoid teleoperation main body is one layer in the slave or master end, not the entire chain. The upper body and wheeled chassis are a modular structure. How to connect to other chassis after disassembly is confirmed per the project interface; it is not written in the public table as already connected. A multi-degree-of-freedom dexterous hand or gripper can be added to the end effector according to the task. The page states compatibility with grippers and dexterous hands from Aoyi, Qiangnao, Zhixing, and others, as well as the RX75, RM75, and RM65 robotic arm platforms. Compatibility is not a standard-configuration list. If a person records their own hand motion using a glove or head-mounted device, that is human motion, not joint commands for this device. Both recordings can be done simultaneously, but the file roles must be kept separate.
In the published methods, ALOHA places master–slave dual-arm teaching and block-wise learning of motion in a single pipeline; Mobile ALOHA extends the problem to a mobile base; UMI deliberately uses a handheld gripper rather than a whole machine. These works show that task definition changes the action space. They did not test the wheeled humanoid teleoperation main body, and the success rates and durations in the papers cannot be written into the task specification.
Therefore, the deliverable of this article is not a "recommended purchase," but a task boundary: only evaluate operations that fall within the above published range; for heights, masses, mobility, or end effectors outside the range, open a separate configuration row, or switch to a different product card. The entry point remains the Product Center, and the parameter page is /product-body-humanoid.html.

How to Use This Page in This Version
Write the task boundary as the first paragraph of the experiment plan, not in the appendix. The first paragraph may only cite sentences already published about the wheeled humanoid teleoperation robotic platform, and must not write subjectless statements like "platform capabilities cover common manipulation". Verify height with this sentence: vertical reach 2100 mm. Verify horizontal reach with this sentence: horizontal reach 1818 mm, excluding the end effector. Verify mass with this sentence: single-arm payload 5 kg. Total machine weight is approximately 125 kg, not single-arm payload. Verify ground operation with this sentence: movement speed 0.1–1.5 m/s. Turning radius 690 mm, measured in place. Only when all three sentences are covered may the task proceed to cameras and personnel; if any one sentence is not covered, stop at the first paragraph and revise the task or switch product pages.
The second paragraph explains where the motion comes from. The end effector can be equipped with a multi-degree-of-freedom dexterous hand or gripper depending on the task. The page states compatibility with Aoyi, Qiangnao, Zhixing and other grippers and dexterous hands, as well as the RX75, RM75, RM65 robotic arm platforms. Compatibility is not a standard configuration list. The upper body and wheeled chassis are modular structures. How to connect to other chassis after disassembly is confirmed per project interface, and is not written in the public table as already connected. If this paragraph requires a primary arm, open the kit page and write the column name; do not count the primary arm in the price and weight of the wheeled humanoid teleoperation platform. If this paragraph requires a gripper, write the gripper model; do not treat "quick-change" or "compatible" as a model. A person wearing a headset or gloves only indicates how the human moves; do not write it as a robot action in the file name.
The third paragraph states what not to do. Do not write charging of approximately 2 hours as runtime of 2 hours. Do not write the folded version's 20 DoF or approximately 6 hours onto this machine. Do not write the lift version's 900 mm as this machine's lift stroke. These negative statements are placed on the same page as the task sentences, so that a reviewer does not have to reach the end of the document to discover that the hours or degrees of freedom were borrowed from an adjacent page. Negative statements do not disparage the device; they circumscribe the acceptable range.
The fourth paragraph should describe sample expectations. For an initial evaluation, first collect trajectories that can be replayed: joint or master–slave correspondence, open/close actions, and whether the chassis moves. Do not commit to success rates before the task boundaries are frozen. Success rates published in papers belong to the original authors' robot, task, and success definition; they must not be presented as expectations after using a wheeled humanoid teleoperation platform.
The fifth paragraph states the exit conditions. Stop capture and change the configuration if any of the following occurs: the object mass exceeds the rated load and there is no written overload test; the operating height is not within the defined vertical range or reach; the required end effector is not listed on the configuration sheet; the required battery life in hours does not exist in the power row. Exiting is not failure — it prevents non-comparable trajectories from being written into the same training set.
Line-by-Line Verification
Degrees of freedom. Total active degrees of freedom 21, broken down as dual arms 14, waist 3, legs 2, neck 2. This row only states the number and locations of joints that have already been published. It does not evaluate intelligence, and it does not automatically include the waist, legs, neck, or whole-machine totals that are not named on the page. When writing into the action vector, create channels only for the body parts listed in this row; body parts not listed remain blank. Do not write blank as 0, because 0 would be treated by the training script as a real angle. Do not fill in the degrees-of-freedom integer from another device to complete this row. If the master–slave joint order is not given on the page, mark it as "to be confirmed in written configuration"; do not number it yourself. The task specification only promises that these published joints can be recorded; it does not promise joints outside the page. The action vector creates channels only for the body parts named in this row; body parts not named remain blank, not 0.
Load and mass. Single-arm load: 5 kg. Overall machine weight: approximately 125 kg — this is not the single-arm load. Every number in this row must be read with its subject: rated, self-weight, overall machine mass, envelope, friction, or gripping force. If the subjects differ, they must not be merged into a single number, and no conversion not given on the page may be performed. If field objects are heavier, the specification is not changed to match; conversely, a larger specification does not prove that every grasping method reaches that limit. The tool mass is undetermined — do not write the arm's rated load as the graspable object mass. The acceptance table retains the original units. If an object exceeds the rated or recommended load in this row, the task is changed to a different object, not the specification. During acceptance, read the subject first, then the number; if the subjects are inconsistent, split into two rows.
Form and scale. Overall dimensions 570 × 465 × 1680 mm. Chassis width and shoulder width are kept within 465 mm. This line describes form or structural scale, used to check 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 on this line, note separately that they are not part of the specification. If weight appears in an adjacent row, do not include it in the dimension sentence. Doors and elevators at the task site must pass this line first; if they do not, do not schedule that task for this device. Doors, elevators, and transport are checked against form only — do not treat this line as operating height.
Operating range. Vertical arm reach: 2100 mm. Horizontal arm reach: 1818 mm, excluding the end effector. This row describes the range the hand or lift can reach; the definition must follow the original wording: vertical working range, arm reach, motion radius, or stroke — these cannot be added together or renamed. If "excluding the end effector" is stated, it must remain in the figure caption. If the original text does not specify units, retain the note "units not specified in the original" — do not delete it for tidiness. If the operating height or depth falls outside the range defined in this row, it is not a default task for this equipment. The figure caption retains the range definition from the original wording, as well as the terms "excluding end" or "units not specified."
Motion. Motion speed: 0.1–1.5 m/s. Turning radius: 690 mm, specified as in-place rotation. This row describes how the ground platform or main arm moves. Four-wheel steering, four-wheel drive, differential drive, in-place rotation radius, and rotational speed are terms for different mechanisms and cannot be substituted for one another. Published speed is a capability, not a recommended teaching speed; teaching may be slower, but the slower speed is recorded under experimental conditions and is not written back as a specification. If the page does not state gradeability or positioning, do not borrow those figures from another page. Whether the task requires moving the chassis or the main arm, first check whether the mechanism terms in this row match. Teaching speed may be lower, but the lower speed is recorded under operating conditions and does not change the published capability in this row.
Power. Battery 40 Ah (BMS). Charging time approximately 2 hours. Approximately 2 hours is charging time, not battery life; the page does not state battery life in hours. This line retains only the electrical terms that appear on the page: voltage, capacity, battery life, charging, current, or power. If there are no "hours," write that battery life is not given — do not fill in hours from an adjacent page. Do not rename charging time as battery life. Rated and maximum values must be listed separately. Keep "approximately" for approximate values. Do not sum the ampere-hours of multiple devices and present them as one device's battery. Task duration can only be written into the plan if hours or charging are stated on this line. Hours, charging, or current that do not appear remain as not given — do not copy them in from an adjacent page.
Perception and computing. Vision system: binocular depth / RGB, optional. Wide-angle cameras: 130° FOV × 3, located at chest / back / chassis. LiDAR: single-line laser, detection range greater than 40 m (at 90% reflectivity). Voice interaction module: ring-shaped 6-microphone array, optional. Depth camera versions: data-collection edition Aobi 335LG; standard edition Intel RealSense D405; optional Intel RealSense D435. This row lists the items explicitly stated for sensors, computing, or communications, and retains the terms standard, included, optional, and optional-order. An optional item that is not selected does not count. If the cover conflicts with the list, cite the list while not pretending the cover statement does not exist. Headsets and gloves do not belong in this row. Processors, storage capacity, or camera channel counts that are not stated must not be drawn into the system diagram. The number of task video channels is governed by this row, not by verbally added cameras. Optional sensors that have not been purchased do not enter this data dictionary.
End effector. A multi-degree-of-freedom dexterous hand or gripper can be added to the end according to the task. The page states compatibility with grippers and dexterous hands from Aoyi, BrainCo, Zhixing, and others, as well as the RX75, RM75, and RM65 robotic arm platforms. Compatibility is not a standard-included list. This row describes where contact or gripping occurs. Quick-change, compatible, and recommended are not models already in the box. The ring gripper on the main arm is not an end gripper. Force, stroke, and load remain as stated in their respective rows and are not merged into a single system metric because they are installed on a particular body. Flange and protocol mappings that are not provided are marked "to be confirmed," not "plug and play." Grasping tasks must specify the end-effector model separately and cannot stop at the interface description in this row. "Supports" or "compatible" without a model number cannot be treated as delivered.
Boundary. The upper body and wheeled chassis are a modular structure. How to connect after disassembly to other chassis is confirmed per the project interface; do not write it in the public table as already connected. 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 prevents verbally buying two pages as one. When writing the proposal, copy the boundary into "not included in this order." If equipment not included is added later, open a new configuration sheet; do not go back and change the platform name in old samples. When a task depends on equipment on the other side of the boundary, add a procurement line to the plan. Equipment within the boundary requires a separate order; do not rename old samples to say it is already included.
Configuration column. The voice module and depth camera versions each have optional rows. Optional items are not written into the deliverables unless checked. The end-effector brand must be named explicitly in the configuration sheet; do not write "compatible means included." This column distinguishes standard, recommended, and optional. Recommended and optional items are not delivered unless checked. When a standard row conflicts with a parameter table row, 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 sign-off. Price and lead time are not written into this column's restatement. Check this column only after the task review has passed; unchecked items do not enter the experiment. Unchecked recommended and optional items are written on the order as not delivered.
Excerptable. The following may be copied into an appendix: 21 (14+3+2+2); 570×465×1680 mm; approximately 125 kg; 465 mm width limit; vertical arm reach 2100 mm; horizontal arm reach 1818 mm (excluding end); in-place rotation radius 690 mm; 0.1–1.5 m/s; single arm 5 kg; 40 Ah; charging approximately 2 hours. This row is the type of content that can be pasted into an appendix: it carries a subject, carries units, and can be traced back to the original sentence on the product page. When pasting, retain "approximately," "excluding," "not provided," "optional-order," and "recommended." Do not convert and keep only the result. If the same number has different meanings on two pages, write it as two rows in the appendix. Test protocols must not be mixed into specification excerpts. After the task is frozen, paste the excerptable original sentences into the appendix to prevent the meeting minutes from altering units. When pasting into the appendix, paste the words "approximately," "excluding," and "optional-order" along with the numbers.
Do not write. Do not describe a charge time of approximately 2 hours as a battery life of 2 hours. Do not attribute the foldable model's 20 DoF or approximately 6 hours to this unit. Do not describe the height-adjustable model's 900 mm as this unit's height-adjustment travel. This line is a negative statement for review purposes. It does not describe what the device can do; it only prevents hours, degrees of freedom, payload subjects, or cover slogans from adjacent pages from being written in. If a negative statement is matched, revert to the original product page sentence and do not interpret it as "close enough." Do not write in success rates from external papers or total hours not published on this site. Before finalizing the task brief, search using the negative statements; if any are matched, return the draft for rewriting. If the manuscript search matches a negative statement, revert to the original product page sentence.
| Projects | Public-Facing Positioning of This Page |
|---|---|
| Degrees of Freedom | Total active degrees of freedom: 21, broken down as dual arms 14, waist 3, legs 2, neck 2. |
| Load and Mass | Single-arm payload 5 kg. Total unit weight is approximately 125 kg, which is not the single-arm payload. |
| Form Factor and Scale | Overall dimensions 570 × 465 × 1680 mm. The chassis width and shoulder width are kept within 465 mm. |
| Operating Range | Vertical reach 2100 mm. Horizontal reach 1818 mm, excluding the end effector. |
| Motion | Motion speed 0.1–1.5 m/s. Turning radius 690 mm; the specification refers to in-place rotation. |
| Power | Battery 40 Ah (BMS). Charging time approximately 2 hours. The approximately 2 hours refers to charging, not runtime; the page does not state runtime in hours. |
| Perception & Compute | The vision system is listed as binocular depth / RGB optional. Wide-angle camera 130° FOV ×3, located on the chest / back / chassis. The LiDAR is a single-line laser with detection range greater than 40 m (90% reflectivity). The voice interaction module is a ring-shaped 6-microphone array, optional. Depth camera versions: data collection version Orbbec 335LG; standard version Intel RealSense D405; optional Intel RealSense D435. |
| End Effector | A multi-degree-of-freedom dexterous hand or gripper can be added to the end effector depending on the task. The page explicitly states compatibility with grippers and dexterous hands from Aoyi, Qiangnao, Zhixing, and others, as well as the RX75, RM75, and RM65 robotic arm platforms. Compatibility is not a standard configuration list. |
| Boundary | The upper body and wheeled chassis are modular. How to connect to other chassis after separation must be confirmed per the project interface; do not state in the public table that it is already connected. |
| Configuration Panel | Both the voice module and depth camera versions have optional rows. Optional items are not included in unchecked deliveries. The end-effector brand must be specified by name in the configuration sheet; do not write "compatible means already included." |
| Excerptable | Attachment copy: 21 (14+3+2+2); 570×465×1680 mm; approximately 125 kg; 465 mm maximum width; vertical reach 2100 mm; horizontal reach 1818 mm (excluding end-effector); in-place rotation radius 690 mm; 0.1–1.5 m/s; single arm 5 kg; 40 Ah; approximately 2 hours of charging. |
| Do not write | Do not write "approximately 2 hours of charging" as "2 hours of battery life." Do not attribute the folding model's 20 DoF or approximately 6 hours of battery life to this unit. Do not write the lifting model's 900 mm as this unit's lifting stroke. |

Relationship to Public Methods and Adjacent Pages
Teleoperation data capture requires separating three layers first, rather than first choosing an "all-in-one platform." The first layer is the master side where the operator issues actions — it can be an isomorphic master arm, a glove, or a controller. The second layer is the slave-side rig that executes the actions — this is the equipment discussed in this article series. The third layer is the recorded data written to disk: timestamps, master-side commands, slave-side joint or chassis state, images, and end-effector open/close state. When the three layers come from different pages, the order, acceptance, and training set must all be written separately.
Published work has already explained this split clearly. Zhao et al.'s ALOHA places low-cost dual-arm master-slave teleoperation and Action Chunking with Transformers in the same pipeline: the human demonstrates through the master arm, the slave arm executes, and the policy learns action chunks consistent with that slave arm. The later Mobile ALOHA puts dual-arm demonstration on a mobile base, discussing how the task distribution changes after mobility, rather than copying the numbers from a fixed base to a mobile base. Chi et al.'s Diffusion Policy shows that demonstrations can train visuomotor policies, but the action representation must match the robot used during capture. Chi et al.'s UMI uses a hand-mounted gripper to capture in-the-wild manipulation, explicitly not an entire humanoid or mobile dual-arm. Open X-Embodiment combines data from multiple robots, on the premise that embodiment differences are labeled, rather than averaging parameters from different embodiments into a single row.
These references only provide methodological positioning. They have not tested any embodiment sold by Hunan TOBI Technology. The success rates, data durations, and camera configurations in the papers cannot be written up as TOBI's own measured results. What TOBI's product pages publish are device boundaries and specification tables. Egocentric video captured by a person wearing a head-mounted device or a glove is also not an action label for this embodiment.
So the "task boundary" article answers only one question: which classes of operator actions can this base unit handle, and which actions the product page never promised. Once the boundary is clear, degrees of freedom, payload, cameras, and battery will not be read as marketing adjectives.
Before handing off to a colleague
Before handing it off to a colleague, read the task boundary out loud.
The first sentence retains only the original definition of height or range, the second sentence retains only the load subject, and the third sentence retains only the mobility mechanism. Do not connect the three sentences with "and the combined capability is sufficient". Then check whether the end device and the master device are already on separate lines; if not, write "end device and master device not included" on the cover of the task sheet. Finally, check for hours or degrees of freedom borrowed from a neighboring page. Delete the borrowed sentences, and only then can the task sheet enter review. The review does not discuss model architecture; it only discusses whether these three sentences cover the site. If they do not cover it, change the product page or change the task; do not change numbers during the review. The second page of the task sheet contains only failure conditions, not vision. Failure conditions include: height not within the original sentence's definition, object mass without a corresponding subject, chassis or main arm mechanism terms that do not match the site, no model number for the end device, no column name for the master device, hours written into the schedule that are not in the power row. After each failure condition, write "stop and change the configuration", not "collect a little first and see". "Collect a little first and see" will put non-comparable trajectories into the same directory, and they will be difficult to separate later. The review signature signs only this page of failure conditions, not a paragraph describing capabilities.
First, the task specification cover must not contain capability adjectives without a subject.
Second, height, load, and mobility each get their own row.
Third, the end device and the master device either have a model number or are written as not included.
Fourth, hours and degrees of freedom from adjacent pages do not appear.
Fifth, the reviewer signs only the failure conditions.
Sixth, if it fails, change the configuration first — do not start collecting.
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.