ML Engineer, Manipulation

Role overview Build and deploy machine learning systems that let mobile robots manipulate objects reliably in changing, human-centered environments. The work spans sensor-driven models, training data and evaluation, simulation-to-real transfer, and deployment on robot hardware. You will use field results to improve reliability, safety, and performance across real-world tasks.

Responsibilities - Develop learning-based models for robot interaction, including reaching, motion generation, and action execution. - Create training pipelines using robot logs or teleoperation data, including dataset preparation, action representations, augmentation, and distributed training. - Define evaluation metrics and regression checks for reliability, recovery behavior, and safety. - Build simulation-to-real workflows with domain randomization and failure-mode testing. - Optimize and distill models for edge hardware, measuring latency, memory use, and stability. - Integrate learned policies with control and safety systems, then validate performance on robots. - Analyze field failures and improve models through targeted data collection and retraining.

Requirements - Bachelor’s or master’s degree in robotics, computer science, electrical engineering, or a related field. - At least three years applying machine learning to robotic manipulation, visuomotor control, or sequence models. - Strong PyTorch skills and experience building dependable training and evaluation pipelines. - Strong Python software engineering skills and ability to work across ML and robotics teams.

Nice to have - Experience with vision-language-action models, behavior cloning, or transformer and diffusion policies for robotic control. - Simulation-to-real manipulation training experience, including synthetic data and domain randomization. - Edge deployment experience, such as ONNX or TensorRT, quantization, or performance profiling. - Familiarity with safety-critical robotics and fallback or recovery behaviors.