Wetour Robotics | Austin, TX (Onsite) | Project-Based | Paid | Apr-May 2026
Intro
Wetour Robotics builds the OS layer for Physical AI. We are launching Orchestra, our first public product, at an industry event in Austin in mid-May 2026. This role owns the neural map that translates raw muscle signals into real-time hand intent. What you build will run live on stage at the launch.
What You'll Do
- Build the sEMG-to-Kinematic labeling pipeline using MediaPipe for ground-truth skeletal generation.
- Develop and train deep temporal models (Transformer/GRU/CNN) for continuous 21-node hand skeletal regression from non-linear muscle potentials.
- Deploy and optimize inference engines using TensorRT and Quantization-Aware Training (QAT) to hit under 20ms end-to-end latency on Jetson hardware.
What We're Looking For
- Advanced proficiency in PyTorch with hands‑on experience in time‑series or biological signal modeling.
- Expertise in ONNX optimization and edge‑device model deployment.
- Familiarity with biomechanical constraints and skeletal rigging logic.
- Ability to view sEMG signals as a high‑fidelity map of human intent, not noise to be filtered out.
- Obsessed with latency — 20ms is the ceiling, not the target.
- Comfortable at the intersection of biological signals and skeletal rigging logic without needing a manual.
- Experience deploying TensorRT and QAT by default, not after everything else fails.
- Passion evident through a personal folder of hand‑tracking experiments.
- Fluent in English, written and spoken.
Role Details
Project-based internship, now through mid‑May 2026. Preferred In‑person in Austin, TX. Paid.
#J-18808-Ljbffr