Build the rig
Pi 5 control, sensor integration, steady flow, and reference-air switching. Establish a repeatable baseline.
A nose for robots.
We’re building a small sensor module that draws in air and records how its gas sensors respond.
View the hardwareProposed sensing channelPrototype in development
4 × BME688 · ENS160 · SGP41
Test odors.
First, we need to check whether the sensors can distinguish odor sources across different days.
We’ll collect labeled samples under controlled conditions and record temperature and humidity alongside each run.
How it
samples air.
A pump, gas sensors, and a reference-air valve form the proposed sampling system.
Draw in air
A pump draws air through a narrow sensing channel. A sampling valve controls the pulses of air entering the channel.
Pump + sampling valveProposed air path / schematic
Inside the first test rig
- Controller
- Raspberry Pi 5
- Gas sensing
- 4 × BME688, ENS160, SGP41
- Ambient pod
- SCD-41 CO₂ + PMS5003 particles
- Flow channel
- 12 cm aluminum channel; approximately 9 mL
- Sampling design
- Steady pump with valve-controlled sniffing
- Current stage
- Test-rig development and validation
These are design choices for the first rig. Detection speed, classification accuracy, and long-term stability still need experimental validation.
What it
records.
The planned dataset pairs gas readings with sampling events, room conditions, and camera frames on one clock.
Shared timestampsSimulated data
Gas response
Raw sensor readings and heater steps
Room conditions
Temperature, humidity, CO₂, particles
Sampling events
Pump, valve, and reference-air states
Camera frames
Frame IDs on the same clock
73747576Example visualization only. No measured detection or classification result is shown.
Example record
Each record will include the sensor ID, a timestamp, the reading, and the sampling state.
Download a simulated record to inspect the proposed format. The data interface may change after testing.
Simulated values / proposed JSON formatThe first
test rig.
We’re two engineers working on the first rig. These are the next steps.
Test across days
Labeled odor sessions across multiple days. Check performance on a day the model hasn’t seen.
Connect to a robot
Use the test results to revise the hardware and define a robot-facing data interface.
Parts and cost
The design uses off-the-shelf sensors and a Raspberry Pi 5. Cost estimates are provisional. There is no retail price or shipping date.
Contact
Smel.
Tell us what your robot needs to detect and where it will operate.
Prototype development
Research inquiries
Research inquiries