Applied Robotics Engineer

Medal.tv; New York (NY); On-site; USD 140,000 - 190,000; Full time

Job summary

Medal.tv is seeking an Applied Robotics Engineer to bridge research with real‑world AI deployment on hardware, navigating constraints such as limited compute, power, and noisy sensors. The role includes post‑training evaluation, system integration, and deployment, with field work alongside partners in various locations.

You will own the process end‑to‑end, from research to running models on hardware, debugging failures, and optimizing inference under real‑world limits.

Qualifications

• 5+ years deploying AI or robotics systems from prototype to production.
• Experience across hardware‑software boundary: sensors, robotics middleware, edge compute, real‑time constraints, and unreliable networks.
• Inference optimization on constrained hardware: quantization, distillation, TensorRT or ONNX export, control frequency.

Responsibilities

• Bridge research with deployment on real hardware under constraints like limited compute and noisy sensors.
• Work with partners to understand legacy control systems, sensor pipelines, latency constraints, and edge compute limits.
• Own the end-to-end process from model checkpoint to running on hardware, including debugging and root-cause analysis.

Skills

Python

PyTorch

C++

Sensor fusion

Edge compute

Real-time constraints

Robotics

Tools

TensorRT

ONNX

Job description

• We are looking for an Applied Robotics Engineer to bridge our research with the realities of deploying AI systems in the physical world


• Our models are trained on real human decisions in simulated environments, and your job is to make them work on hardware in the real world, under constraints such as limited compute, power, noisy sensors, and unpredictable environments. You will take our models and focus on post-training, evaluation, system integration, and deployment


• You will work directly with partners to understand their legacy control systems, sensor pipelines, latency constraints, and edge compute limits. This position requires both office and field work, with time working closely with customers in various locations


• You’ll also close the loop back to research. When a policy fails on hardware you’ll find out why, whether it’s sensor calibration, actuator dynamics the simulator didn’t capture, control-loop timing, coordinate conventions, or a model leaning on one modality and ignoring another


• This is an in-person role based in our New York City HQ

Comfortable with a partner’s engineers and their leadership in the same room, explaining what a model will and won’t do without overselling itYou have taken a model from a training checkpoint to running on real hardware and owned it end to end, including the failuresExperience across the hardware-software boundary: sensors, robotics middleware, edge compute, real-time constraints, and unreliable networksStrong Python and PyTorch, and you drop into C++ when performance requires it. You’ve done inference optimization on constrained hardware: quantization, distillation, TensorRT or ONNX export, control frequencyYou have debugged a model that was quietly wrong rather than obviously broken, found the root cause, and fixed the system rather than the symptomDepth in at least one of the following and interest in the others: policy learning and control (RL, imitation learning); world models and simulation (Unity, Unreal, Isaac); perception and inverse dynamics models (sensor fusion, learning from demonstration)

5+ years of experience deploying AI or robotics systems from prototype to production, spanning simulation, model development, and real-world deployment

You’re rigorous about evaluation, and you know what a benchmark number does and doesn’t tell you