ML Engineer II, Manipulation (Robotics)

TL;DR

ML Engineer II, Manipulation (Robotics): Develop and deploy learning-based manipulation systems that enable mobile robots to interact reliably with the physical world in dynamic human environments with an accent on perception-to-action models, training datasets, evaluation tooling, and deployment pipelines. Focus on building sim-to-real workflows, optimizing models for edge deployment, and driving iterative improvements through field performance analysis and targeted retraining.

Location: Anywhere in the US

Company

We build artificial intelligence that enables service robots to collaborate with people and adapt to dynamic human environments.

What you will do

• Develop learning-based manipulation models for end-to-end sensor-driven interactions like reaching, motion generation, and execution in dynamic environments.
• Build and maintain manipulation training pipelines including dataset creation, action representations, augmentation, and distributed training.
• Design evaluation metrics and regression tests to quantify reliability, recovery behavior, and safety in real environments.
• Develop sim-to-real workflows with simulation environments, domain randomization, and failure-mode testing.
• Optimize and distill models for edge deployment, benchmarking latency, memory, and stability on target hardware.
• Partner with AI platform team to integrate policies with control and safety systems, validating end-to-end performance on robots.
• Analyze field performance, identify failure modes, and drive improvements via data collection and retraining.

Requirements

• Bachelor’s or Master’s in Robotics, Computer Science, Electrical Engineering, or related (PhD a plus)
• 3+ years applying ML to robotics manipulation, visuomotor control, or sequential models
• Strong proficiency in PyTorch and building reliable training/evaluation pipelines
• Strong software engineering skills in Python; ability to collaborate across ML and robotics teams

Nice to have

• Experience with Vision-Language-Action (VLA) models, behavior cloning, transformer/diffusion policies for robotic control
• Sim-to-real training for manipulation (Isaac Sim/Mujoco), domain randomization, synthetic data
• Deploying ML models to edge hardware (ONNX/TensorRT, quantization, performance profiling)
• Familiarity with safety-critical robotics integration and fallback/recovery behaviors