Python SDK
The definitive execution engine for data-heavy spatial applications. Process massive mimetic datasets, run cloud-scale biomechanical evaluations, and power real-time Sim-to-Real translation bridges with native physics access.
pip install willow-runtime
The Analytics and Retargeting Engine
While the C++ SDK is designed for embedded hardware triggers, the Python SDK is built for deep spatial analytics and robotic integration. It bridges the gap between raw 3D coordinate streams and actionable physics. Use this SDK to query the Cloud Oracle, pull down highly complex action models into ephemeral RAM, and stream local inference data directly into high-fidelity physics evaluators (like the Da Vinci Engine) or translate it via native Coordinate Bridges for digital twin simulation.
High-Fidelity Physics Evaluator
Access the core math of the Da Vinci Engine locally. Upon action detection, calculate precise values for topological path efficiency, localized Jerk, and dynamic Peak Power output (kW).
Sim-to-Real Retargeting
Automate human-to-robot motion translation. Leverage native coordinate bridges to map human spatial telemetry perfectly to ROS (Z-Up) and Nvidia Isaac Sim environments for RL training.
Longitudinal Data Aggregation
Process batches of human movement to calculate Mechanical Consistency Scoring. Track micro-deviations in angular velocity and hip-shoulder separation across thousands of repetitions.
Zero-Shot Semantic Search
Scan massive, unstructured S3 video lakes or sensor archives for specific kinetic events. Find the exact frame an ergonomic fault or specific robotic gesture occurs without retraining a neural net.
Scale-Invariant Execution
Our proprietary torso-length normalization guarantees that an action model trained on a 6'5" subject evaluates the motion of a 5'2" user flawlessly, enabling universal physics baselines.
DRM RAM-Only Security
Designed for highly secure enterprise, defense, and healthcare deployments. Proprietary kinematic models are loaded directly into ephemeral memory over HTTPS, protecting intellectual property.
from willow import WillowClient, WillowDetector, CoordinateBridge, KinematicRetargeter
# 1. Load Model (Secure RAM Stream)
client = WillowClient(api_url="https://api.willowdynamics.com/v1", api_key="...", customer_id="...")
model = client.get_model("tactical-reload-v1")
detector = WillowDetector(model)
# 2. Continuous Semantic Search & Retargeting Loop
while True:
current_skeleton, timestamp_ms = get_next_frame()
# .step() executes in <2ms
event = detector.step(current_skeleton, timestamp_ms)
if event:
print(f"Action Detected at {event['end_ms']}ms | Confidence: {event['confidence']:.2f}")
# 3. Translate MediaPipe (Y-Down) to ROS/Isaac Sim (Z-Up)
ros_ready_sequence = CoordinateBridge.to_ros_z_up(current_skeleton)
# 4. Extract Joint Angles for RL Training or Physics Logging
angles = KinematicRetargeter.extract_joint_angles(ros_ready_sequence)
print(f"Lead Elbow Flexion: {angles['left_elbow_flexion']}")