You've received a dexterous hand sample for evaluation. The next seven days will determine whether it earns a place in your robot platform or goes back in the box. Most teams evaluate samples by "playing with it" for a few days and making a gut-feel decision. That approach misses critical failure modes and leads to costly late-stage discoveries.
This guide provides a structured 7-day evaluation protocol used by our engineering team when qualifying new dexterous hands for the RobX product line. Each day has a specific objective, a checklist, and a go/no-go gate. If a hand fails at any gate, you stop — there's no point continuing with a fundamentally flawed component.
1. Day 1: Unboxing and Physical Inspection
Before connecting any power or communication cables, inspect the hand physically. Mechanical issues discovered here save you hours of debugging later.
Day 1 Checklist
Weigh the hand — compare against datasheet spec (±5% tolerance)
Measure dimensions — verify against datasheet drawings
Inspect all finger joints for free movement — no binding, no grinding
Check cable connectors — verify pin count, orientation, locking mechanism
Inspect for shipping damage — bent fingers, cracked housings, loose screws
Photograph all sides — establish baseline condition for warranty
Verify included accessories — power supply, cables, USB adapter, mounting bracket
Read the quick-start guide — note any red flags (missing steps, unclear diagrams)
| Check | Pass Criteria | Fail = Stop |
|---|---|---|
| Weight vs. datasheet | Within ±5% | >10% deviation suggests wrong configuration |
| Joint free movement | All joints move smoothly by hand (power off) | Any joint that binds or grinds = mechanical defect |
| Connector condition | Clean, no bent pins, locks positively | Bent pins or loose fit = unreliable connection |
| Documentation quality | Quick-start guide with clear wiring diagram | No documentation = integration will be painful |
2. Day 2: Communication and Basic Control
Today's goal: establish bidirectional communication and command a single joint to move. If you can't talk to the hand by end of Day 2, integration will be a long battle.
Step-by-Step
Identify the communication protocol — CAN, CAN-FD, EtherCAT, or serial. Check the datasheet and connector pinout
Set up the bus interface — USB-to-CAN adapter, EtherCAT master, or serial-USB converter
Install drivers and SDK — Follow manufacturer instructions. Note: if instructions are only in Chinese, flag this as a support risk
Power on the hand — Use the specified voltage (typically 24V DC). Watch for LED indicators
Send a ping/heartbeat command — Verify the hand responds with status data
Read current joint positions — Confirm all DOF report valid angle values
Command a single joint to move — Start with a small movement (10°). Gradually increase
Test emergency stop — Verify you can halt motion instantly
Day 2 Checklist
Communication established (protocol identified and working)
All DOF report valid position data (no NaN, no frozen values)
Single joint movement command succeeds
All joints respond to commands (test each one)
Emergency stop works reliably
SDK/API documentation is usable (code examples run)
Temperature readings are within normal range during operation
| Common Day 2 Problems | Likely Cause | Resolution |
|---|---|---|
| No response from hand | Wrong CAN baud rate or node ID | Check datasheet defaults, try 1000 kbps / 500 kbps |
| Some joints respond, others don't | Addressing conflict or daisy-chain break | Check wiring continuity, verify node IDs |
| Jittery or erratic motion | Control gains not tuned, or noise on bus | Check grounding, try lower PD gains |
| SDK examples crash | Library version mismatch, OS compatibility | Contact supplier for working version |
| Overheating at idle | Firmware bug or wiring issue | Stop immediately — hardware risk |
3. Day 3: Motion Characterization and Calibration
Now that the hand moves, you need to understand exactly how it moves. Motion characterization is the foundation for all downstream control work.
Key Measurements
| Parameter | How to Measure | Compare Against |
|---|---|---|
| Joint range of motion (ROM) | Command each joint from min to max, measure actual angle with protractor or camera | Datasheet ROM spec |
| Position accuracy | Command 5 target angles per joint, measure actual vs commanded | ±0.5° typical for good hands |
| Repeatability | Command same angle 10 times, measure variation | ±0.1° for precision hands |
| Maximum speed | Command max velocity, time a full flex-extension cycle | Datasheet speed spec |
| Step response time | Command a step input, measure time to 90% of target | <100 ms typical |
| Backlash | Move to position from both directions, measure position difference | <0.5° for harmonic drive, <2° for tendon |
Fingertip Position Mapping
For applications requiring precise manipulation, you need to map joint angles to Cartesian fingertip positions. This forward kinematics model is essential for grasp planning and trajectory generation.
3D-print or machine a calibration grid (known reference points)
Command the hand to touch each grid point
Record joint angles at each point
Fit a DH-parameter or learned kinematic model
Verify model accuracy at 5 random points not used for fitting
4. Day 4: Grasp Testing and Force Measurement
Today's goal: determine what the hand can actually hold, and with how much force. This is where theoretical specs meet physical reality.
Test Object Set
Prepare a standardized set of test objects covering different grasp types:
| Grasp Type | Test Objects | What You're Testing |
|---|---|---|
| Power grasp (cylindrical) | Water bottle, can, cylinder (50-80mm dia.) | Maximum grip force, hand wrap-around capability |
| Precision grasp (pinch) | Pencil, pen, small cylinder (5-15mm dia.) | Fingertip dexterity, fine force control |
| Lateral pinch | Credit card, thin book, phone | Thumb-to-finger lateral force |
| Tripod grasp | Pen (writing position), small ball | Three-finger coordination |
| Hook grasp | Bag handle, rope, briefcase handle | Load-bearing capacity, finger strength |
| Delicate grasp | Egg, ripe fruit, paper cup | Force regulation, compliance, tactile feedback |
Force Measurement
If you have a force gauge (or even a kitchen scale), measure:
Maximum fingertip force — command maximum grip, measure force at fingertip
Force resolution — command 5 force levels (10%, 25%, 50%, 75%, 100%), measure actual
Force repeatability — command same force 10 times, measure variation
Slip detection (if tactile sensors present) — grip an object, gradually increase load until slip occurs, verify sensor detects it
Day 4 Checklist
All 6 grasp types tested with appropriate objects
Maximum grip force measured and compared to datasheet
Force control resolution acceptable (can hold delicate objects without crushing)
Slip detection working (if applicable)
Grasp success rate >80% on standard test objects
No dropped objects during 10-repeat grasp cycle
5. Day 5: Durability and Thermal Testing
A hand that works perfectly for 5 minutes but degrades after 30 minutes of continuous use is useless for real applications. Day 5 is about finding the thermal and fatigue limits.
Continuous Operation Test
Program a repetitive grasp cycle: open → grasp standard object → release → repeat
Run continuously for 2 hours minimum
Log: joint temperatures every 15 minutes, grasp success rate, position accuracy drift
Stop if any motor temperature exceeds 70°C (or manufacturer-specified limit)
| Metric | Good | Acceptable | Failing |
|---|---|---|---|
| Motor temperature after 1 hour | <45°C | 45-60°C | >60°C (thermal shutdown risk) |
| Grasp success rate (start vs 1 hour) | No degradation | <10% drop | >10% drop (durability concern) |
| Position accuracy drift | <0.5° drift | 0.5-1° drift | >1° drift (thermal expansion or encoder drift) |
| Audible noise change | No change | Slight change | New grinding/clicking = mechanical wear |
Cycle Count Estimation
Run 1,000 grasp cycles and measure performance degradation. If performance is stable after 1,000 cycles, the hand is likely good for 10,000-50,000 cycles before significant wear. If degradation appears before 1,000 cycles, expect frequent maintenance.
6. Day 6: Integration with Your Robot Platform
Day 6 is where the hand stops being a standalone device and becomes part of your robot. This is also the day that reveals integration headaches.
Integration Tasks
Mount the hand — Attach to your robot's wrist flange. Verify weight doesn't exceed arm payload, cable routing is clean, and no collisions during wrist rotation
Route communication cable — Through the arm's internal cable channel. Check for pinching at joint extremes
Power integration — Connect to robot's power bus (verify voltage stability under load)
Software integration — Implement the hand's driver in your robot's control framework (ROS2, custom, etc.)
Coordinate frame calibration — Define the hand's TCP (tool center point) relative to the arm's flange frame
End-to-end test — Command the arm to move to a target, have the hand grasp an object, move the arm away
Control Architecture Check
Verify your control architecture handles the hand properly:
Control loop runs at the required frequency (1 kHz for position, 100+ Hz for grasp planning)
Safety watchdog works — if communication drops, hand opens (not freezes closed)
Emergency stop halts both arm and hand simultaneously
No interference between arm and hand control threads
7. Day 7: Documentation Review and Go/No-Go Decision
Day 7 is synthesis day. Review everything you've learned, score the hand, and make a decision.
Evaluation Scorecard
| Category | Weight | Score (1-5) | Weighted Score |
|---|---|---|---|
| Mechanical quality (build, joints, materials) | 20% | __/5 | __ |
| Communication reliability (driver stability, no dropouts) | 15% | __/5 | __ |
| Motion performance (accuracy, speed, repeatability) | 20% | __/5 | __ |
| Grasp capability (force, object variety, success rate) | 15% | __/5 | __ |
| Durability (thermal, cycle life, wear) | 10% | __/5 | __ |
| Documentation quality (SDK, English docs, examples) | 10% | __/5 | __ |
| Integration effort (time from sample to working on robot) | 10% | __/5 | __ |
| Total | 100% | __/5.0 |
Decision Matrix
| Total Score | Decision | Action |
|---|---|---|
| 4.0 - 5.0 | GO | Proceed to procurement. Order production units + spare parts |
| 3.0 - 3.9 | CONDITIONAL GO | Acceptable with caveats. Address specific weaknesses before scaling |
| 2.0 - 2.9 | HOLD | Significant issues. Request manufacturer fixes or evaluate alternatives |
| <2.0 | NO-GO | Reject. Return sample and look for alternatives |
8. Evaluation Scorecard Template
Here's a printable checklist you can use for each sample evaluation:
Complete 7-Day Evaluation Checklist
Day 1: Physical inspection passed, weight/dimensions within spec
Day 1: No shipping damage, all accessories included
Day 1: Quick-start documentation available and clear
Day 2: Communication established, all DOF responding
Day 2: SDK examples run without modification
Day 2: Emergency stop verified
Day 3: ROM within spec for all joints
Day 3: Position accuracy within ±0.5°
Day 3: Repeatability within ±0.1°
Day 4: All 6 grasp types tested successfully
Day 4: Maximum grip force within 80% of datasheet spec
Day 4: Delicate objects held without damage
Day 5: 2-hour continuous operation — temp <60°C
Day 5: 1,000 cycles — no performance degradation
Day 6: Mounted on robot, no cable issues
Day 6: End-to-end grasp test successful
Day 6: Safety watchdog verified
Day 7: Scorecard completed, total score calculated
Day 7: Go/No-Go decision documented with rationale
Frequently Asked Questions
Can I compress this into 3 days?
You can, but you'll skip durability testing (Day 5) and platform integration (Day 6). These are the two areas where hidden problems surface. We've seen hands that pass Day 1-4 perfectly but fail catastrophically after 30 minutes of continuous use. Compressing the protocol risks missing these failure modes.
What if the hand doesn't come with an SDK?
No SDK is a major red flag. You'll need to reverse-engineer the communication protocol, which adds weeks to integration and makes future firmware updates a gamble. At minimum, require a communication protocol specification document and sample code.
Should I evaluate multiple hands simultaneously?
If you have the engineering bandwidth, yes. Parallel evaluation lets you compare hands directly and makes trade-offs visible. However, each hand needs its own dedicated engineer for the full 7 days — don't split one person across two hands.
What's the most commonly missed failure mode?
Thermal management. Many hands work perfectly for 5-10 minutes but overheat during continuous operation. Day 5's 2-hour continuous test catches this — and it's the test most teams skip because "the hand works fine on the bench."
How much should evaluation cost?
One engineer × 7 days × $1,200/day = ~$8,400 in evaluation labor. This sounds expensive, but it's a fraction of the cost of integrating the wrong hand for 3 months and then switching. Treat sample evaluation as an investment, not an expense.
Need a Sample Evaluation Kit?
RobX Robotics provides evaluation support — including test object sets, measurement guidance, and US-based engineering assistance during your 7-day evaluation.
Request Evaluation Support