Founding Edge Intelligence Lead | Robotics & Actuation

About the job Location: New York City preferred Role type: Full-time Seniority: Senior / Principal / Founding Engineer Compensation: Competitive cash compensation, benefits, and founding-engineer level equity package. This is one of the company’s first technical hires and will carry meaningful ownership, technical authority, and long-term upside tied to building a foundational part of the Sentor platform. About the company The company is building , a patent-pending AI-native motion platform for robotics and complex electromechanical systems. The AI-native motion platform is designed to move intelligence closer to the physical motion layer. Instead of treating actuation as a passive output from a central controller, we are building hardware and software that can sense local conditions, adapt to changing loads, and respond in real time. We recently closed our seed round and are building the founding technical team. The Role We are looking for a Founding Edge Intelligence Lead to build the onboard machine learning and decision-making layer for the AI-native motion platform. This role sits at the intersection of machine learning, robotics, controls, embedded systems, and physical hardware. Your job is to turn signals such as load changes, vibration patterns, motion signatures, thermal behavior, and other physical signals into useful intelligence: state estimation, anomaly detection, adaptive response, and eventually learned motion behavior. This is not a cloud ML role. It is not a dashboard role. It is not a generic AI role. It is the software layer that lets a physical machine learn from its own body. What You’ll Own 1. Onboard Sensing and State Estimation You will develop models and algorithms that interpret Sentor’s local sensor data in real time. This may include: torque resistance estimation; load and contact inference; vibration and resonance mode prediction and detection; friction / hysteresis pattern recognition; thermal and mechanical drift prediction and detection; fatigue / degradation signatures; sensor fusion across current, position, velocity, torque, temperature, strain, vibration, and related signals. The goal is to help AI-native motion platform to understand what is happening locally before waiting for a centralized robot controller. 2. Edge ML and Local Decision-Making You will build the onboard intelligence layer that enables AI-native motion platform to respond intelligently to changing physical conditions. This may include: lightweight ML models for low-power embedded inference; adaptive torque / speed / stiffness response; local reflex-like behaviors; learned response profiles; anomaly detection and protective behaviors; real-time classification of operating regimes; prediction of near-term load or resistance changes; safe local adaptation within guardrails. The system must be fast, robust, and suitable for real-time actuation. Model elegance matters less than useful, reliable behavior on hardware. 3. Controls-Aware Learning This role must bridge ML and controls. You will work on algorithms that can coexist with, augment, or inform control systems rather than fight them. Relevant areas may include: model predictive control; adaptive control; reinforcement learning for constrained physical systems; imitation learning; system identification; state-space models; Kalman filtering / particle filtering; residual learning around physics-based models; learned compensators for friction, hysteresis, backlash, vibration, or load variation. The right person understands that onboard intelligence in an actuator needs to be safe, bounded, and physically grounded. 4. Simulation and Digital Twin Interface You will work closely with our simulation / digital twin function. Over time, the edge intelligence layer should interface with simulation tools so we can: generate synthetic operating scenarios; test edge models before hardware deployment; simulate load changes, impacts, failures, and edge cases; train or validate models in virtual environments; compare simulated sensor traces with real bench data; accelerate sim-to-real iteration; develop motion profiles before physical customer deployments. The long-term vision is for AI-native motion platform onboard intelligence and AI-native motion platform simulation environment to reinforce each other. 5. Embedded Deployment You will help turn ML models into deployable real-time software. This will include: embedded C model development and deployment model compression and optimization; embedded inference; latency and memory optimization; real-time data pipelines; firmware / embedded software integration; safety fallbacks; deterministic behavior under constrained compute; deployment to microcontrollers, embedded Linux, DSPs, or edge accelerators. The job is not done when the model works in Python. The job is done when it improves behavior on the machine. Ideal Background Strong candidates have experience with: embedded C code; robotics; autonomous systems; humanoids / legged robots; drones; surgical robotics; industrial automation; motor control; embedded ML/edge AI; controls; physical AI; sensor fusion; reinforcement learning for real systems; predictive maintenance / anomaly detection for machines; simulation-to-real robotics. Relevant experience may include: time-series ML; real-time sensor fusion; embedded inference; controls-aware ML; actuator or motor control; adaptive systems; state estimation; robot dynamics; online learning or calibration; physics-informed ML; digital twins; sim-to-real workflows. Tools and Technical Stack Preferred experience: Embedded C; Python, C++; PyTorch, JAX, TensorFlow, scikit-learn; ONNX, TensorRT, TFLite Micro, TVM or related deployment tools; MATLAB / Simulink; ROS / ROS 2; MuJoCo, Isaac Sim, Gazebo, Drake, or related robotics simulation tools; real-time data acquisition; embedded Linux, RTOS, microcontrollers, DSPs, or edge accelerators; control systems, MPC, state-space modeling, filtering, and system identification. We care less about exact tool overlap and more about whether you can build useful intelligence for real physical systems. What Success Looks Like In the first 3–6 months, success looks like: building the first sensor data pipeline for Sentor; defining the onboard intelligence architecture; identifying which sensor signals are most predictive of physical state; creating initial models for load / resistance / operating-regime inference; deploying early edge models or algorithms on prototype hardware; building safety-bounded local response behaviors; integrating with bench-test data and simulation outputs; helping define how Sentor learns from repeated physical interaction. Longer term, this role becomes foundational to Sentor’s differentiation: local intelligence at the actuator level, coordinated across the robot through Sentor Motion Studio. You Are a Fit If You are excited by ML that touches real hardware. You think physical intelligence should live closer to the body, not only in the cloud or central controller. You can move between Python experiments, embedded C code, control theory, embedded constraints, and physical test data. You understand that real-time systems require safety, reliability, and bounded behavior. You are comfortable building with incomplete data and improving through iteration. You want to help create the onboard intelligence layer for a new category of actuation.