ML Architect for Robotics Infrastructure at Stealth Startup.

About the role

We’re a small, stealth robotics company working on high-resolution tactile sensing for manipulation. Our hardware produces dense, image-like signals from contact with the world, and we need someone to turn that raw data into clean, calibrated tensors that downstream learning and control systems can actually use.

In this role, you’ll define the internal “tactile tensor” representations, build the calibration and compensation models, and work closely with an embedded/driver engineer who owns low-level drivers and data streaming.

What you’ll do

• **Architect the representation
  **Design the tactile tensor layouts used internally for learning and control (dimensions, units, metadata, coordinate frames, sparse vs dense structures).

• **Physics-aware calibration
  **Build ML and signal-processing pipelines that map raw sensor outputs (images, voltages, capacitance, etc.) to grounded physical quantities such as forces, deformation, and contact geometry.

• **Model the messy real world
  **Implement models to handle thermal drift, hysteresis, wear, and other real-world artifacts, turning noisy streams into stable, scientific-grade measurements.

• **Close the hardware loop
  **Work closely with an embedded/driver engineer to:

• Define logging schemas and APIs between drivers and ML components.

• Debug data issues (timing, sync, dropped frames, saturation, coordinate misalignment).

• Distinguish whether a glitch is electrical, mechanical, or algorithmic.

• **Bridge simulation and reality
  **Analyze and align data from physics engines (e.g., Isaac Sim, MuJoCo, PyBullet) and real hardware to quantify and reduce the sim-to-real gap.

• **Design experiments and benchmarks
  **Design and run experiments end-to-end: data collection, training, evaluation, and reporting. Help specify small benchmarks and demo tasks (e.g., grasp stability, slip detection) that showcase tactile capabilities.

Requirements

Education / background

• Advanced degree (MS/PhD) in Robotics, Computer Science, Electrical Engineering, Applied Physics, or a related field or equivalent practical experience building ML/robotics systems.

Strong Python fluency and experience with modern ML frameworks

• PyTorch preferred; TensorFlow or JAX also fine.

• Solid background in machine learning plus at least one of:

• Computer vision (convolutions, geometry, calibration)

• Robotics / robot learning

• Sensor fusion, control, or time-series modeling

Experience with real hardware / robots / sensors

• You don’t have to write firmware, but you are comfortable working with an embedded engineer, inspecting logs and plots, and forming hypotheses about what’s wrong.
• You can reason about whether an issue is coming from electronics, mechanics, or the model.

Signal processing & calibration intuition

• Experience mapping raw signals (images, voltages, capacitance, etc.) to physical units and validating those mappings experimentally.
• Comfort with physical units, coordinate transforms, and calibration math.

Simulation exposure

• Familiarity with physics engines (Isaac Sim, MuJoCo, PyBullet, etc.) and how contact/friction are modeled, or a willingness to get up to speed quickly.

Communication rigor

• Ability to write clear technical specs, experiment reports, and occasional math-heavy docs, and to explain why certain architectures or approaches were chosen.

Nice to have

• PhD or research background in robotics, tactile sensing, sim-to-real, or related areas.
• Background in tactile sensing, haptics, or robotic manipulation.
• Experience with optical tactile sensors (e.g., GelSight-style, DIGIT-style) or other high-dimensional force sensing.
• Experience with sim-to-real transfer techniques (domain randomization, system identification, etc.).
• Exposure to metrology or other precision measurement work.
• Some C++ or systems knowledge, enough to understand constraints on the driver side.