Paxini PX-6AX GEN3 Multidimensional Tactile Sensor

1. Overview

The PaXini PX-6AX GEN3 is a family of high-resolution multidimensional tactile sensors designed for robotic fingertips, palms, and finger pads. The "6AX" designation denotes the sensor's ability to resolve contact forces and torques along all six spatial axes (Fx, Fy, Fz and Tx, Ty, Tz), going far beyond single-axis load cells or simple binary contact switches. This makes the GEN3 suitable for contact-rich manipulation tasks such as object stiffness estimation, grasp stability monitoring, slip detection, and high-fidelity tactile data collection for imitation learning.

The GEN3 series ships as individual sensor tiles in three anatomical form factors — fingertip (DP), finger pad (IP/CP), and palm (MC) — across several size variants. Each tile mounts to a compatible robotic hand or end effector, connects to one of three communication board options (SPI hub, serial converter, or high-speed integrated board), and streams data over USB to the host PC. The PXSR software suite provides real-time visualization, raw data export, and coordinate-mapped signal display using the per-variant XLSX taxel layout files included in the product bundle.

Typical users are robotic manipulation researchers integrating tactile feedback into dexterous hand platforms, and data collection teams building contact-rich demonstration datasets for behavior cloning and reinforcement learning pipelines.

2. Full Specifications

Sensor Variants

Every variant is identified by a part number in the format PX6AX-GEN3-{category}-{size}-{tier} and ships with a corresponding XLSX signal-coordinate file named PXSR-STD{category}03{suffix}.xlsx.

Communication Boards

Three communication board configurations are available. Choose based on the number of sensors and required data throughput:

3. Quick Start

From unboxing to live data in PXSR:

1. Mount the sensor and connect the communication board. Attach the GEN3 sensor tile to your robot fingertip, finger pad, or palm mount. Connect the sensor ribbon to the appropriate communication board. Plug the communication board USB cable into your Windows x64 host.
2. Install the PXSR software. Run pxsr-gen3-win-x64-1.0.7_Release.exe from the 【04】 Software Installation Package folder. The installer registers the USB driver and places the PXSR executable in C:\Program Files\PaXini\PXSR\.
3. Load the variant XLSX coordinate file. Open PXSR. Under File → Open Coordinate Map, select the XLSX file matching your sensor variant (e.g., PXSR-STDDP03B.xlsx for the 26mm fingertip). This maps raw sample indices to physical taxel positions on the sensor surface.
4. Connect and visualize. Click Connect in PXSR. The live tactile heatmap renders immediately. Apply gentle pressure to the sensor surface to confirm signal response. Force/torque bars on the right panel show the 6-axis decomposition in real time.
5. Export data. Use Record → Start to capture a session as a CSV or binary log. Each row contains a Unix timestamp, the 6-axis F/T values, and the full raw taxel array.

4. SDK / Python Usage

The GEN3 product bundle includes three C reference implementations for reading sensor data directly over the raw interface, bypassing PXSR entirely. These serve as the foundation for custom integrations on Linux, embedded systems, or real-time control loops.

UART Example

For the single-channel serial converter board. Configure the serial port at the baud rate specified in the communication protocol document, then read frames using the packet structure defined therein:

/* UART Example.c — included in 【02】Code Example 通信例程 */
/* Open serial port, set baud rate per protocol doc, read frames */
int fd = open("/dev/ttyUSB0", O_RDWR | O_NOCTTY);
struct termios tty;
tcgetattr(fd, &tty);
cfsetispeed(&tty, B115200);   /* confirm rate from protocol doc */
cfsetospeed(&tty, B115200);
tcsetattr(fd, TCSANOW, &tty);
/* Read frame, parse 6-axis F/T values and taxel array */

Python wrapper (platform bridge)

For integration with the Fearless Platform, wrap the C UART reader in a Python subprocess or use pyserial directly. The following skeleton matches the JSONL telemetry format expected by the platform WebSocket:

import serial, json, time

ser = serial.Serial("/dev/ttyUSB0", baudrate=115200, timeout=0.1)

def read_frame(ser):
    # Parse according to PaXini GEN3 protocol doc v1.0.5
    raw = ser.read(64)           # frame size varies by board; see protocol doc
    ft = parse_force_torque(raw)  # your parser here
    taxels = parse_taxel_array(raw)
    return {"ts": time.time(), "ft": ft, "taxels": taxels}

while True:
    frame = read_frame(ser)
    print(json.dumps(frame), flush=True)  # emit JSONL for platform bridge

SPI multi-sensor (10-channel hub)

When using the 10-channel SPI hub, each channel is polled round-robin. The SPI Example.c in the product bundle demonstrates the select-and-read pattern. With 10 sensors at typical sample rates you can achieve full-hand tactile coverage at 100+ Hz aggregate throughput, sufficient for contact-onset detection and slip events in manipulation tasks.

5. Platform Integration

The Fearless Platform at platform.roboticscenter.ai supports tactile sensor streams as first-class nodes in a teleop session. The GEN3 sensor data is surfaced in the GloveWorkbench and TeleopCameraStreamsPanel UI components, which render real-time tactile heatmaps alongside joint-state and camera streams from the same session.

Registration sequence

1. Start a teleop session via POST /api/teleop/sessions/create-sdk with device_type: "paxini_gen3".
2. Connect the sensor bridge (Python script above) to the session WebSocket at /api/teleop/ws.
3. Send the registration handshake: role: "robot", device_type: "tactile_sensor", capabilities: ["tactile", "force_torque", "telemetry"].
4. On receiving type: "ready", begin streaming telemetry frames. Each frame should include type: "telemetry", the 6-axis F/T values, the flattened taxel array, and a Unix millisecond timestamp.

Telemetry frame format

{
  "type": "telemetry",
  "member_id": "paxini-fingertip-right",
  "device": "paxini_gen3",
  "variant": "PX6AX-GEN3-DP-M2826-Omega",
  "ft": {
    "fx": 0.12, "fy": -0.04, "fz": 1.83,
    "tx": 0.001, "ty": -0.002, "tz": 0.0
  },
  "taxels": [0.0, 0.02, 0.15, ...],  // flat array, length = variant taxel count
  "ts": 1743680400123
}

Episode recording

During an active recording session, all telemetry frames from the GEN3 sensor are archived as JSONL alongside hand joint positions and camera frames. The platform episode browser allows playback of the tactile stream with the spatial heatmap rendered using the variant XLSX coordinate map. Episodes are accessible for download or direct use in training pipeline APIs.

6. Troubleshooting

7. Related Hardware

The PaXini GEN3 sensor is commonly used alongside dexterous hand platforms. See the following wiki pages for integration guidance on the hands themselves:

• LinkerBot O6 — 5-finger dexterous hand with glove controller; the GEN3 13mm fingertip variants are dimensioned for O6 finger mounting.
• Wuji Dexterous Hand — Wuji hand with 5×4 joint position sensing; GEN3 tactile tiles can be added for contact feedback.
• VLAI L1 — Full humanoid robot; palm M2020 and fingertip variants are applicable to its end effectors.

Also see the PaXini Tactile Sensor Resources page for additional technical documentation, integration notes, and the Robotics Academy tactile sensing learning path.