Machine Learning Engineer, Perception R&D

Responsibilities:

• Build Performance Analysis Systems: You will design systems that automatically analyze robot telemetry and video logs to detect anomalies, quantify perception failures, and provide deep insights into why a robot performed a certain way.
• Scale the Data Engine: You will take ownership of the data lifecycle. You will build automated pipelines to mine interesting events from fleet data, implement active learning strategies to curate high-value datasets, and drive the continuous improvement of our model suite.
• Develop Scene Understanding Models: You will train and refine 2D/3D vision models specifically designed for event detection and semantic understanding.
• Master Edge Deployment: You will own the optimization and deployment of these models. You will use tools like TensorRT and ONNX to ensure our models run efficiently on edge hardware (NVIDIA Orin).
• Define Perception Metrics: You will move beyond simple validation loss and define real-world performance metrics that correlate with customer success, visualizing these trends for the broader engineering team.

Skills:

• 3+ Years of Experience in Computer Vision and Machine Learning, with a strong focus on MLOps, deployment, or data infrastructure.
• Strong proficiency in Python and PyTorch. You are comfortable writing tooling scripts, data queries, and deployment pipelines.
• You have experience optimizing models for the edge using TensorRT, ONNX Runtime, or CUDA. You understand how to profile model performance and manage resource constraints.
• You treat data as a product. You have experience utilizing SQL, data lakes, or vector databases to manage large-scale datasets and automate quality assurance.
• Familiarity with VLMs and 2D/3D Vision architectures sufficient to debug model failures and understand data needs.
• You can translate high-level product requirements into specific engineering tasks and explain technical trade-offs to non-expert stakeholders.
• Familiarity with Docker, AWS/GCP (S3, EC2), labeling platforms and experiment tracking tools.