TouchGlove SDK

High-performance Python SDK for Touch Glove 5-channel tactile data acquisition, RTC time synchronization, real-time 3D displacement/force physical inference (ONNX Runtime / OpenVINO), and H.265 video recording with microsecond-level timestamps.

Powered by a native Rust core with high-efficiency PyO3 bindings.

⚡ Quick Start

Installation

pip install imprintx

Basic Streaming Example

from imprintx import TouchGlove, list_ports
import time

# Scan for available serial ports
ports = list_ports()
print("Available ports:", ports)

if not ports:
    raise RuntimeError("No serial ports found!")

# Connect to glove device and start streaming
with TouchGlove(port=ports[0]) as glove:
    print("Device SN:", glove.get_sn())
    glove.sync_rtc()  # Synchronize host clock with hardware RTC
    glove.start()

    print("Streaming tactile data... Press Ctrl+C to stop.")
    while True:
        batch = glove.poll()
        for frame in batch:
            print(f"Ch {frame.channel} | Seq: {frame.seq_id} | Timestamp: {frame.timestamp_us} us | Image: {frame.image.shape}")
        time.sleep(0.01)

✨ Features

• High-Throughput 5-Channel Streaming: Concurrent acquisition of 192x192 8-bit tactile images across 5 channels.
• Hardware RTC Synchronization: Precision clock alignment between host OS and glove hardware RTC.
• Real-Time Physical Inference: ONNX Runtime and OpenVINO inference for 3D displacement fields (32, 32, 3) and 3D force fields (32, 32, 3) with auto baseline calibration.
• H.265 Video Recording & Microsecond Timestamps: High-efficiency HEVC grid video encoding + microsecond timestamp JSON export.
• Cross-Platform Support: Native binaries for Linux (x86_64, aarch64), Windows, and macOS.

📖 Key Python APIs

1. TouchGlove Class

Main interface for device management, data acquisition, and inference.

glove = TouchGlove(
    port="/dev/ttyACM0",     # Serial port path
    model="dense_3ch.onnx", # Path to ONNX model (optional)
    device="auto",           # "auto", "cuda", "gpu", "npu", "cpu", "mac"
    auto_open=True           # Auto handshake on initialization
)

Methods

• Device Management:

  • glove.open(port: str) / glove.close(): Manage serial connection.
  • glove.start() / glove.stop(): Start / stop tactile data stream.
  • glove.poll() -> FrameBatch: Fetch latest batch of 5-channel tactile frames.
  • glove.is_open() -> bool / glove.is_streaming() -> bool: Query connection status.

• Baseline Calibration & Inference:

  • glove.calibrate_baseline(duration_sec=1.0): Capture baseline and calibrate zero point for live inference.
  • glove.get_dense_fields(): Get latest 5-channel 3D displacement and force fields (5, 32, 32, 3).

• Hardware SN & RTC:

  • glove.get_sn() -> str: Query hardware serial number (SN).
  • glove.set_sn(sn: str) -> bool: Write hardware serial number.
  • glove.sync_rtc() -> bool: Sync host OS system time to glove RTC.
  • glove.query_rtc() -> Optional[str]: Query current glove RTC timestamp string.

• Video Recording:

  • glove.start_recording(output_path, fps=30.0, crf=18): Record 5-channel grid H.265 video.
  • glove.stop_recording() -> str: Stop recording and generate microsecond timestamp JSON file.

2. Frame Class

Represents a single frame from one channel.

• frame.channel: Channel index (0–4).
• frame.sensor_id: Hexadecimal sensor hardware ID.
• frame.seq_id: Frame sequence number.
• frame.timestamp_us: Hardware reception timestamp in microseconds.
• frame.image: Raw tactile image as numpy.ndarray (shape=(192, 192), dtype=uint8).
• frame.disp: Inferred 3D displacement field (shape=(32, 32, 3), dtype=float32, optional).
• frame.force_field: Inferred 3D force field (shape=(32, 32, 3), dtype=float32, optional).

3. FrameBatch Class

Synchronized batch containing frames across all 5 channels.

• batch.frames: List of 5 Frame objects (or None for inactive channels).
• batch.timestamp: Host Unix timestamp (seconds).
• batch.total_frames: Count of valid frames in batch.
• Supports iteration: for frame in batch: ...

🛠 Advanced Examples

Model Inference with Baseline Calibration

from imprintx import TouchGlove, list_ports
import time

ports = list_ports()
with TouchGlove(port=ports[0], model="dense_3ch.onnx", device="cuda") as glove:
    glove.start()
    
    print("Calibrating baseline for 1 second...")
    glove.calibrate_baseline(duration_sec=1.0)
    print("Calibration complete!")
    
    while True:
        batch = glove.poll()
        for frame in batch:
            if frame.disp is not None and frame.force_field is not None:
                print(f"Ch {frame.channel} Max Force: {frame.force_field.max():.2f}")
        time.sleep(0.01)

Video Recording & Timestamp Query

from imprintx import TouchGlove, list_ports, load_video_timestamps
import time

ports = list_ports()
with TouchGlove(port=ports[0]) as glove:
    glove.start()
    glove.start_recording("tactile_recording.mp4", fps=30)
    time.sleep(5.0)  # Record 5 seconds
    glove.stop_recording()

# Parse exported microsecond timestamps
records = load_video_timestamps("tactile_recording.mp4.json")
print(f"Total recorded frames: {len(records)}")