PaXini Tactile Sensor Technical Documentation
PaXini PX-6AX GEN3 Technical Documentation
1. System Overview
The PaXini PX-6AX GEN3 series is a family of multidimensional tactile sensing modules designed to provide spatially distributed, triaxial force measurements for robotic manipulation and contact-rich interaction.
Each module integrates sensing, signal processing, force reconstruction algorithms, and digital communication into a compact unit, enabling direct integration into learning-based robotic systems.
The sensor outputs time-synchronized force fields, rather than single-point force values, making it suitable for tactile perception, manipulation learning, and multimodal data collection.
2. Sensing Principle
2.1 Physical Architecture
PX-6AX GEN3 adopts a semi-flexible, multi-layer structure:
Elastic deformation layer
Deforms under contact and transmits force information.
Rigid sensing layer
Hosts Hall-effect sensors that detect magnetic field changes.
Embedded magnetic elements
Translate mechanical deformation into measurable magnetic displacement.
Force-induced deformation changes the relative position of magnetic sensing points, which are then reconstructed into triaxial force vectors (Fx, Fy, Fz) via onboard algorithms.
This architecture enables:
Measurement of normal and tangential forces
High durability under repeated contact
Stable sensing under real-world conditions
2.2 Magnetic Interference Mitigation
To ensure signal integrity, PX-6AX GEN3 includes anti-stray-field compensation algorithms that suppress interference from:
External magnetic fields
Strong lighting conditions
Environmental electromagnetic noise
This is critical for long-duration data collection and deployment in non-laboratory environments.
3. Sensor Output Model
3.1 Distributed Force Representation
At each sampling step, the sensor reports:
Per-cell triaxial force vectors
\mathbf{f}_i = (F_{x,i}, F_{y,i}, F_{z,i})
Aggregated resultant forces
\mathbf{F} = \sum_i \mathbf{f}_i
Force units:
Resolution: 0.1 N per LSB
Normal force (Fz): 0–25 N
Tangential force (Fx, Fy): ±10 N
Example:
A reported Fz value of 10 corresponds to 1.0 N normal force.
3.2 Spatial Resolution
Spatial resolution: 1 mm
Minimum detectable force: 0.1 N
Accuracy: 1% FS
Depending on the module variant, sensing arrays range from 9 to 239 measurement points, enabling high-density tactile perception at fingertips, finger pads, or palms.
4. Coordinate System
Each sensor defines a local global coordinate frame anchored at a fixed reference point on the sensor housing.
All sensing points are expressed in this frame
Triaxial force axes are consistent across modules
Sensing point positions are fixed and known
This explicit geometric grounding allows tactile data to be:
Aligned with robot kinematics
Fused with vision and proprioception
Used directly as structured learning observations
5. Hardware Variants
The PX-6AX GEN3 family includes Elite, Core, and Omega variants, differing in:
Physical dimensions
Number of sensing points
Maximum current consumption
All variants share:
Identical sensing principles
Consistent force scaling
Unified communication semantics
This enables cross-device dataset consistency.
6. Electrical Characteristics
Operating voltage: 3–5 V DC
Logic level: 3.3 V
Maximum current:
Elite: up to 150 mA
Core: up to 350 mA
Omega: up to 700 mA
Sensors are designed for embedded integration and can be powered from standard robotic control boards.
7. Communication Interfaces
7.1 Supported Protocols
PX-6AX GEN3 supports multiple digital interfaces:
Protocol
Characteristics
Use Case
SPI
High throughput, deterministic timing
Real-time learning and control
UART
High baud rate (921600)
Embedded systems, debugging
I²C
≤200 kHz
Low-speed integration
Protocol selection occurs only at power-on, based on pin configuration.
7.2 Addressing and Module Identification
Device address = module index + 1
Multiple sensors can be daisy-chained
Chip-select pins encode module role (DP / IP / CP / Palm)
This enables scalable deployment on multi-fingered robotic hands.
8. Register Map and Data Access
8.1 Address Spaces
The sensor exposes two main register regions:
User Configuration Area (0x79)
Calibration trigger
Control flags
Write-limited (non-contiguous writes only)
Application Area (0x7B)
Read-only force data
Continuous streaming supported
Per-cell force values stored sequentially
8.2 Force Data Layout
Force data is stored as:
[Fx1, Fy1, Fz1, Fx2, Fy2, Fz2, ...]Each value:
1 byte
Little-endian
Scaled at 0.1 N per unit
This layout is optimized for:
DMA transfer
Real-time streaming
Efficient tensor conversion
9. Data Acquisition Workflow
9.1 Calibration
Calibration must be performed under no-load conditions
Triggered via configuration register
Zero-offset applied internally
Proper calibration is required before any data collection session.
9.2 Streaming and Logging
The host software supports:
Real-time force visualization
Continuous data logging
Timestamp-aligned recording
This enables direct ingestion into:
Dataset pipelines
Learning frameworks
Offline analysis tools
10. Learning-Oriented Data Semantics
From a learning system perspective, PX-6AX GEN3 provides:
Contact localization (where touch occurs)
Force magnitude and direction
Shear and slip indicators
Temporal evolution of contact
These signals are critical for:
Grasp stability detection
Slip prediction
Contact-aware control policies
Failure and recovery modeling
11. Safety and Operating Limits
The sensor is designed for:
Touching
Gripping
Squeezing
Sliding contact
It is not intended for:
Piercing or cutting contact
Exposure to fire or corrosive liquids
Strong magnetic fields
Medical, military, or aviation systems
Safe load:
200% rated load
300% impact overload
12. Maintenance and Reliability
Service life: >10 million contact cycles
Elastomer surface can be cleaned with:
Adhesive tape
Lint-free cloth
Mild alcohol (low concentration)
Organic solvents and aggressive chemicals must be avoided.
13. Intended Integration Scenarios
PX-6AX GEN3 is intended for:
Learning-based robotic manipulation
Tactile-enhanced imitation learning
Reinforcement learning with contact feedback
Multimodal dataset creation (vision + touch + proprioception)
It is explicitly designed to be a data source, not just a sensor.
14. Summary
The PaXini PX-6AX GEN3 series provides dense, spatially grounded, triaxial tactile data through a robust, digitally accessible interface.
By exposing tactile perception in a form suitable for learning systems and dataset pipelines, PX-6AX GEN3 enables robots to reason about contact—not as an event, but as a measurable, learnable signal.