Strategy 1: Characterize, Then Act¶
In the V3 review
Literature scenarios that measure first and then act (Fig. 3):
- flow → measure → infer: mass-flow viscometry up to 5000 cP [22].
- dispense → measure → generalize: proxy viscometry on a standard Opentrons platform, 20–20,000 cP [23].
- sense → estimate → act: viscosity estimated from stirring forces before planning a pour, calibrated to one batter [31].
- measure → calibrate → dispense: a powder simulator calibrated from measured flowability, generalizing to unseen cohesive powders [39].
The vendor baseline is itself a characterize-then-act lookup table: Opentrons' documented settings reach ~1,000–1,400 cP for a few named liquids [21]. See Literature Scenarios.
Overview¶
Measure material properties once, compute optimal handling parameters, execute.
This is the simplest strategy: take time upfront to understand the material, then dispense or handle it with confidence using calculated parameters.
When to Use¶
- First-time use of a new material or formulation
- Material properties are documented or can be quickly measured
- High confidence in physical model (e.g., Newtonian flow model for a simple liquid)
- Time budget allows for characterization phase before main task
- Variability is low between batches or runs
How It Works¶
Step 1: Measure Material Properties¶
Choose a property that governs handling. Examples:
- Viscosity (for liquids): Use an Anton Paar, TA Instruments, or Brookfield rheometer
- Yield stress (for pastes): Rheometer in controlled stress or controlled rate mode
- Particle size (for powders): Laser diffraction (Malvern Mastersizer) or sieve analysis
- Flowability (for powders): Angle of repose, Hausner ratio, powder rheometer
Time typically required: 30 seconds – 5 minutes per material
Step 2: Build a Model¶
Use the measured property to estimate handling parameters:
Example: Viscosity Model for Pipetting¶
Given: target volume V_target, measured viscosity η
Model: air-gap time Δt = f(V_target, η, pump_model)
For air-displacement pipettes, experimental data show:
Δt (seconds) ≈ 0.1 + 0.5 × (η / 1 cP) [for 1 mL, typical aspirate speed]
Measure the viscosity, look it up in a calibration table, set the air-gap time accordingly.
Example: Yield Stress Model for Syringe Extrusion¶
Given: measured yield stress τ_y, syringe barrel diameter d
Model: required extrusion pressure P = τ_y × (d/2) × safety_factor
The robot applies that force and confirms extrusion starts.
Step 3: Execute¶
Use the calculated parameters for the main task: - Set pump speed, pressure, or gripper force - Execute without further feedback - Record the result
Advantages¶
| Aspect | Benefit |
|---|---|
| Speed | No iterative cycles; main task completes in one pass |
| Simplicity | No feedback loops, fewer sensors required |
| Predictability | Deterministic outcome (if model is accurate) |
| Cost | Works with basic sensors (rheometer, balance, etc.) |
Limitations¶
| Challenge | Example |
|---|---|
| Model accuracy | Non-Newtonian fluids may not follow simple V∝Δt model |
| Hidden variables | Temperature, humidity, batch variation not captured |
| Air bubbles | A bubble in a syringe invalidates volume prediction |
| Drift over time | Viscosity changes as solvent evaporates |
| One-shot recovery | If first execution is wrong, material is already dispensed |
Real-World Examples¶
Example 1: Liquid Handling¶
Task: Dispense 10 mL of a chemical solution to a reaction vessel
Procedure:
- Measure viscosity with a rheometer: η = 2.3 cP
- Look up in calibration table: air-gap time = 0.12 seconds for 10 mL
- Program syringe pump to 10 mL at 0.12 s aspirate, standard dispense
- Execute
Outcome: Liquid dispensed. If bubble-free and material stable, volume is accurate (±2%).
Example 2: Gel Extrusion¶
Task: Extrude 0.1 mL of a polymer gel onto a substrate
Procedure:
- Measure yield stress with rheometer: τ_y = 50 Pa
- Syringe diameter d = 3 mm
- Required pressure P = 50 Pa × (3/2) mm × 1.5 (safety) ≈ 450 kPa
- Program air-pressure syringe system to 450 kPa
- Trigger extrusion
Outcome: Gel extrudes and deposits a bead. Volume is consistent if temperature is stable.
Example 3: Powder Feeding¶
Task: Dispense 5 g of a powder
Procedure:
- Measure Hausner ratio H: if H < 1.25, powder is "free-flowing"
- Use auger feeder at standard speed for free-flowing powders
- Dispense 5 g
Outcome: Powder is dispensed. If measured by weight, accuracy depends on flow consistency (±2–5% typical).
Common Pitfalls¶
1. Trusting a Single Measurement¶
Problem: Measured η = 5 cP, but batch-to-batch variation is ±10%.
Solution: Measure multiple batches, or use a looser tolerance on flow time.
2. Forgetting Temperature Dependence¶
Problem: Viscosity of honey changes by 10% per °C.
Outcome: Morning dispense is accurate, afternoon dispense drifts.
Solution: Either measure at target temperature, or switch to real-time correction (Strategy 3) if temperature fluctuates.
3. Assuming Model Transfer¶
Problem: Model trained on water (density 1.0 g/mL) applied to mercury (density 13.6 g/mL).
Outcome: Pressure/force predictions are off by 10×.
Solution: Re-calibrate model per material class.
4. Ignoring Air Bubbles¶
Problem: Viscosity is correct, but a 0.5 mL air bubble entered the syringe.
Outcome: Aspirated 10.5 mL (air + liquid), dispensed 10 mL, but only 9.5 mL is liquid.
Solution: Always aspirate with a bubble-free technique; use deaeration if needed.
Implementation Checklist¶
- Select property to measure (viscosity, yield stress, flowability)
- Build or select a model relating property to handling parameters
- Validate model on 3–5 materials
- Measure target material property
- Calculate handling parameters
- Execute without feedback
- Verify outcome (weigh, measure, image)
- If accuracy insufficient, escalate to Strategy 2 (Iterative Optimization) or Strategy 3 (Real-Time Correction)
Comparison with Other Strategies¶
| Strategy | Speed | Accuracy | Sensor Cost | Complexity |
|---|---|---|---|---|
| Characterize, then act | Very fast | Moderate (±5%) | Low | Low |
| Iterative optimization | Slower | High (±1%) | Medium | Medium |
| Real-time correction | Medium | High (±2%) | Medium | High |
| Feedforward prediction | Very fast | High (±2%) | Medium | High |
| Simulation generalization | Fast | Medium–High | Medium | Very high |
References¶
- Rastegardoost, M. M.; Bao, Z.; Sadr, S.; Abdollahi, N.; Darvish, K.; Clasky, A.; Ohsaka, H.; Le Devedec, F. Self-Driving Laboratories for Formulation: Handling Liquids, Pastes, Powders and Objects, from Robots to Digital Twins and AI Agents. Manuscript in preparation, 2026.
- Key references from the review: [21], [22], [23], [31], [39]. Full list on the References page.
Further reading (from the earlier wiki)¶
- Mezger, T. G. (2020). The Rheology Handbook (5th ed.). Vincentz Network.