Robotics solution guide · Canada · verified August 1, 2026
Build one task.
Prove every layer.
Short answer: choose the Unitree R1 EDU U5 as a development platform when your pilot genuinely needs two five-finger hands, 38 active joint-motor channels, added compute and secondary-development resources. Do not treat that configuration as proof that a manipulation task already works. A defensible pilot names one object, one start pose, one target placement, the exact software and sensing stack, a controlled test zone and measurable pass/fail evidence. Confirm the delivered U5 bill of materials and interfaces in writing, then move from simulation to hand bench tests and only afterward to integrated robot trials.
Configuration is not an outcome. The U5 listing establishes hardware and development-resource claims. It does not establish grasp success, cycle time, repeatability, safety certification, integration effort or performance on your object.
01 · Configuration truth
What the current R1 EDU U5 listing actually confirms
The live SpeedyDrone listing identifies a specific development configuration: all R1 EDU Smart features, two BrainCo Revo 2 Basic five-finger hands and a 100 TOPS computing expansion dock. These are procurement inputs. Each still needs an integration and acceptance condition.
- Two handsThe listing names two BrainCo Revo 2 Basic five-finger dexterous hands. BrainCo documents six active joints and 11 total mechanical degrees of freedom per hand.
- 38 joint-motor DOFSpeedyDrone calculates 26 body joint motors plus six active joints in each hand. This is an actuator count—not a manipulation score and not a claim that every desired grasp is reachable.
- 100 TOPS dockThe listing names a 100 TOPS computing expansion dock. TOPS is a compute-throughput label; it does not by itself establish model compatibility, memory headroom, end-to-end latency or task success.
- Development resourcesHigh- and low-level secondary development, robot models, simulation interfaces and environments such as Isaac Sim are listed. Require exact asset, interface, software and version delivery because “support” is not the same as a ready-to-run task project.
- 2 kg arm maximumThe current listing states a 2 kg maximum single-arm payload, and Unitree says arm load varies greatly with extension posture. Treat 2 kg as a boundary to investigate, not a working-object recommendation or guaranteed value at full reach.
- Basic hand sensingBrainCo's comparison table does not list a tactile module for Revo 2 Basic. Position, velocity and current feedback are documented, but current feedback must not be relabelled as fingertip tactile sensing.
02 · Scope before hardware
Define one manipulation task before requesting a quote
“Pick and place” is not a specification. A useful pilot freezes the object, scene, motion and evidence so every failure has a meaningful label and every successful repetition can be compared.
Configuration
Robot joints + two Basic hands + compute + developer access
Accepted task
Perception + reach + grasp + hold + placement + recovery, repeated under written conditions
Part number or sample, mass, dimensions, material, surface friction, stiffness, centre of mass, fragile areas and allowed contact zones.
Fixture or marked pose, orientation tolerance, background, lighting, neighbouring objects and whether presentation can vary.
Which hand, approach direction, arm posture, grasp family, transfer path, obstacle clearance, speed limits and reset behaviour.
Target fixture or zone, positional and angular tolerance, release condition, object-damage limit and required evidence.
Good first object: use a non-fragile, low-mass object with stable geometry in a controlled fixture. The objective is to expose interface, calibration and repeatability problems—not to begin at the R1 arm's published maximum or with variable clutter.
03 · Integration path
A six-stage path from configuration to evidence
Each stage has an exit criterion. If the hand interface is not stable on the bench, adding whole-body motion will only make the fault harder to isolate.
-
Freeze the task contract
Photograph and measure the real object, fixture, pickup pose, placement zone and work envelope. Write allowed variation and prohibited contact. Choose which events count as a valid attempt and which conditions require a reset.
Exit evidence: signed task sheet, object sample or drawing, scene diagram, acceptance matrix and named task owner.
-
Lock the delivered stack
Make the quotation identify R1 EDU U5, both left and right BrainCo Revo 2 Basic hands, the 100 TOPS dock, camera, controller, batteries and charger, cables, interfaces, manuals, SDK access, robot and hand models, simulator assets and support boundaries. Confirm how payload is accounted for with the hands installed.
BrainCo documents RS485 and CAN FD for Revo 2 Basic. Unitree's public BrainCo hand service demonstrates a serial-to-DDS pattern for a G1 integration; because that repository names G1, it must not be treated as proof of the exact R1 U5 software image. Request the R1-specific package, mapping and supported versions.
Exit evidence: versioned bill of materials and interface-control document for the quoted unit.
-
Build the digital test
Import the exact robot and hand assets, confirm joint names, limits, coordinate frames, collision geometry, command units and hand orientation. Model the object and fixture with measured dimensions rather than a visually similar stand-in. Record simulator, asset and driver versions.
Use simulation to reject impossible approaches, collisions and poor postures. Do not use a clean simulation run as acceptance of friction, compliance, cable behaviour, camera noise or real grasp stability.
Exit evidence: replayable scene, joint-map check and collision-free candidate sequence.
-
Commission each hand on the bench
Before integrated motion, verify left/right device identity, power, communications, joint direction, range, status reporting and conservative commands. BrainCo states that Revo 2 performs position calibration at startup and that an object held during calibration can be dropped; calibration must finish before normal control. Keep the hand empty and the drop zone clear during that procedure.
Test the chosen grasp on the object with a controlled fixture. Because Basic does not list a tactile module, document the sensing and control signal actually used for contact or stall decisions instead of implying tactile feedback.
Exit evidence: calibrated hand log, object-safe grasp limits and repeatable bench grasp/release record.
-
Integrate at conservative limits
Bring up perception and arm motion with no object first, then a fixture-held object, then the complete transfer. Keep people outside the controlled test zone, begin with conservative speed and acceleration limits, validate the stop and communication-loss response, and keep a recovery method available.
The published 2 kg arm maximum changes with extension posture. Test the real object at the real reach and orientation, and obtain a written load interpretation rather than subtracting or assuming the hand's contribution.
Exit evidence: safe, logged end-to-end cycles at the frozen configuration.
-
Run the acceptance protocol
Freeze software, model, parameters, object batch, lighting, fixture, operator role and test count before the run. Log every valid attempt, success, failure class, intervention, reset and object damage event. Separate “task completed” from “completed within the agreed time and tolerance.”
Exit evidence: signed results against pre-agreed thresholds, unresolved exceptions and a change-control rule for re-testing.
04 · Acceptance design
Measure the whole manipulation chain
A grasp that looks convincing in one video is not an accepted pilot. Use criteria that identify where the chain failed and preserve the conditions needed to repeat the result.
| Criterion | Test method | Pass evidence | Boundary to record |
|---|---|---|---|
| Object presentation | Place the named object inside the allowed start-position and orientation tolerance. | System detects or accepts the object and begins without manual pose correction outside the agreed procedure. | Lighting, background, fixture, object batch and permitted pose variation. |
| Reach and approach | Execute the frozen approach with the actual arm posture and scene geometry. | No unplanned contact, joint-limit violation or excursion outside the controlled envelope. | Which arm, wrist/hand orientation, extension, speed, acceleration and clearance. |
| Grasp acquisition | Close the specified fingers using the documented control signal and limit. | Object leaves the pickup support without prohibited contact or visible damage. | Finger command, current/stall rule, calibration state and absence of tactile sensing on Basic. |
| Hold and transfer | Move through the complete path and any agreed dwell. | No drop, slip beyond tolerance, collision or unplanned protective response. | Object mass, centre of mass, arm posture, path, speed and dwell duration. |
| Placement and release | Place into the measured target and release using the defined sequence. | Final pose is within agreed position/orientation tolerance and object remains undamaged. | Fixture tolerance, target geometry, release height and settling rule. |
| Failure response | Introduce the approved recoverable cases, such as absent object or failed grasp confirmation. | Robot stops or resets according to the written state machine without unsafe movement. | Permitted automatic retry count, human intervention and reset authority. |
| Repeatability | Run the pre-agreed number of valid attempts at the frozen configuration. | Success rate, cycle time, intervention rate and damage rate meet thresholds established before testing. | Attempt denominator, excluded trials, software hash, configuration and environmental conditions. |
One transparent metric
Accepted cycles ÷ valid attempted cycles
Define “valid attempt” before the run. Report setup failures, communication faults, safety stops and manual interventions separately; do not quietly remove them from the denominator after seeing the result.
- Configuration record: robot, hands, compute, camera, firmware, SDK, driver, simulator asset and application version.
- Task record: object identifier, measured mass and geometry, start/target tolerance, grasp, path and timing definition.
- Run record: timestamp, operator, scene, valid attempt, result, cycle time, intervention and failure class.
- Safety record: zone check, stop-function test, communications-loss test, observers and recovery procedure.
- Change record: the hardware, software, object or scene changes that trigger partial or full re-acceptance.
05 · Canadian procurement
Put the pilot boundary into the quotation
For a Canadian lab, innovation team or automation group, the quotation should be a configuration-control document—not just a model name. Commercial terms can change, so this guide deliberately does not state price, stock, lead time or delivery.
Delivered hardware
Exact R1 EDU U5 designation, serialised bill of materials, two Revo 2 Basic hands, compute dock, camera, batteries, charger, controller, cables, spares and documentation.
Development entitlement
High- and low-level interface scope, SDK and model access, R1-specific hand integration, simulator assets, supported operating systems, version baseline, examples and update path.
Pilot services
Name what is included in configuration review, commissioning, integration support, remote or onsite assistance, acceptance witnessing, issue escalation and handover. Do not infer services from a product specification.
Institutional controls
Assign an owner for workplace risk review, test-zone safeguards, IT/network approval, camera-data handling, operator authorization, incident records, maintenance and change control.
Unitree warns that humanoid robots have a complex structure and powerful output, instructs users to maintain a sufficient safety distance and says parameters vary by scenario and configuration. The Canadian Centre for Occupational Health and Safety recommends task-based risk assessment across integration, operation and maintenance. A manipulation pilot therefore needs its own controlled zone, trained roles, stop and recovery procedures, guarding decision and site-specific review; purchasing an EDU configuration does not certify a collaborative application.
SpeedyDrone pathways: verify the live Unitree R1 EDU Pro C / U5 listing, browse SpeedyDrone's current collection directory when comparing adjacent platforms, and use the SpeedyDrone contact page for organization and procurement details.
For facility-level context, read How to Build and Budget a Humanoid Robotics Lab for a Canadian University. That article covers whole-lab planning; this guide remains deliberately narrower: one U5 manipulation task and its acceptance evidence.
06 · Buyer questions
Frequently asked questions
What is included in the current SpeedyDrone R1 EDU U5 configuration?
The live listing says R1 EDU U5 includes all R1 EDU Smart features, two BrainCo Revo 2 Basic five-finger dexterous hands and a 100 TOPS computing expansion dock. It also lists high- and low-level secondary development, robot models and simulation interfaces. The quotation should identify the exact delivered bill of materials, software access and versions.
Does 38 DOF mean R1 U5 can perform any dexterous task?
No. The 38 figure is a joint-motor count: 26 body joint motors plus six active joints in each of two hands. Task success also depends on reach, posture, object geometry, friction, sensing, controls, calibration, software, scene conditions and acceptance thresholds.
How are the 38 joint-motor degrees of freedom calculated?
SpeedyDrone's U5 listing calculates 26 body joint-motor degrees of freedom plus six active motorized degrees of freedom for the left hand and six for the right hand. BrainCo separately describes each Revo 2 as having 11 total mechanical degrees of freedom, six of them active.
Is the listed 2 kg single-arm payload a guaranteed working payload?
No. The listing gives 2 kg as the maximum single-arm payload, while Unitree states that maximum arm load varies greatly with arm-extension posture. Validate the real object, hand configuration, reach, orientation, motion and safety margin, and obtain a written interpretation of payload accounting for the delivered system.
Does the R1 EDU U5 listing guarantee a ready-to-run Isaac Sim workflow?
No. The listing says simulation environments such as Isaac Sim are supported, but that does not guarantee that a specific task scene, exact U5 hand asset, driver mapping or compatible version is preconfigured. Put the required robot model, hand model, interface, simulator version, example and support scope in the quotation.
Do the BrainCo Revo 2 Basic hands include tactile sensing?
BrainCo's current comparison table does not list a tactile module for the Revo 2 Basic version. It documents position, velocity and current feedback control. Do not describe current or stall feedback as fingertip tactile sensing; specify the real contact-detection method used by the pilot.
What should a single-object manipulation acceptance test measure?
Measure object detection or presentation, reach and approach, grasp acquisition, hold and transfer, placement tolerance, failure response, repeatability, cycle time, intervention rate and object damage. Freeze the object, scene, software and configuration, and define valid attempts and numerical thresholds before testing.
What should a Canadian buyer put in the R1 EDU U5 quotation?
Specify the exact hardware and accessories, development interfaces and entitlements, software and simulator versions, R1-specific hand integration, documentation, support boundaries, pilot task, acceptance evidence, delivery and commercial terms. Also assign internal owners for workplace safety, IT, privacy, operation and change control.
Verification record
Official sources checked August 1, 2026
- SpeedyDrone Unitree R1 EDU Pro C / U5 listing — current U5 configuration, two Revo 2 Basic hands, six active DOF per hand, 38 joint-motor DOF, 100 TOPS dock, 2 kg single-arm maximum and development-resource claims.
- Unitree R1 official product page — R1 model table, optional dexterous hands for EDU, secondary development, compute options, posture-dependent arm-load caveat, configuration variability and safety warnings.
- BrainCo Revo 2 official parameters — six active and 11 total hand DOF, Basic interfaces and feedback modes, absence of a listed tactile module for Basic, and power-on calibration behaviour.
- Unitree SDK2 official repository — current SDK environment and documentation route.
- Unitree BrainCo hand service repository — public serial-to-DDS pattern and six-joint command mapping for the repository's stated G1 integration; used only as an interface reference, not as proof of an R1 U5 image.
- CCOHS Robots and Cobots guidance — task-based risk assessment, integration and operation considerations, controls, testing and records for Canadian workplaces.
Quote the task—not only the robot
Send SpeedyDrone the object specification, start and target poses, required cycle, environment, interfaces, software and simulator baseline, safety controls and acceptance thresholds. Ask the quotation to name the exact R1 EDU U5 configuration and support boundary. Current pricing, inventory, lead time and delivery should be confirmed at that point.
Request an R1 EDU U5 Pilot Quote
