Machine Learning Engineer, Perception

Path Robotics is building the future of embodied intelligence with AI-driven systems that enable robots to adapt and learn in real-world scenarios. They are seeking passionate Machine Learning Engineers to develop robotic welding solutions, focusing on perception tasks and deep learning. The role involves implementing machine learning algorithms, building data pipelines, and integrating models into production robotics services.

Responsibilities:

• Implement, validate, and iterate on machine learning algorithms for weld perception tasks, including point cloud registration, seam detection, and joint geometry estimation, progressively expanding coverage across joint types and part geometries
• Build and maintain data pipelines for training and evaluating perception models, spanning annotated 3D scan data ingestion, synthetic data generation, and structured dataset management for iterative model improvement
• Run rigorous model evaluation experiments, including failure mode analysis, FP/FN rate characterization, and benchmarking against quantitative registration accuracy thresholds, and communicate findings clearly to guide next steps
• Integrate trained models into production ROS-based robotics services, ensuring low-latency inference and compatibility with deployed cell configurations
• Write clean, well-tested Python code; participate actively in code and experiment reviews; and maintain clear documentation of methods, parameters, and results
• Lead research, development, and production deployment of advanced perception algorithms spanning point cloud registration, seam detection, and real-time in-process tracking across structured light, RGB, and stereo sensors
• Design and lead experiments evaluating state-of-the-art deep learning models, including transformer-based and geometric feature learning architectures
• Design and lead real-time perception systems such as during-weld seam tracking, applying sensor fusion with probabilistic state estimation (e.g., Kalman filtering) to achieve robust weld performance
• Define and own the end-to-end ML lifecycle, from dataset design and annotation strategy through training, benchmarking, and fleet deployment, with clear go/no-go evaluation frameworks
• Architect distributed training and hyperparameter optimization workflows; drive strategy for data acquisition, annotation tooling, and synthetic vs. real scan data usage
• Mentor engineers across levels, providing technical leadership on perception systems and ML methodology

Requirements:

• Master's or Ph.D. in CS, Robotics, or related field (Computer Vision, ML, or Perception); Bachelor's with strong production ML experience also considered
• 3+ years (Experienced) / 7+ years (Senior/Staff) in real-world robotics or industrial ML
• Strong Python fluency and hands-on PyTorch experience, including training, evaluating, and deploying deep learning models in production
• Experience with 3D point cloud data and libraries such as Open3D, including geometric concepts like surface segmentation, spatial queries, and point-wise labeling
• Familiarity with 3D deep learning architectures such as PointNet++, GeoTransformer, or similar transformer-based or graph-based approaches on geometric data
• Comfortable integrating ML models into production robotics services within ROS-based architectures and containerized deployment environments
• Demonstrated track record leading end-to-end ML projects from dataset design through fleet deployment with rigorous go/no-go frameworks
• Experience architecting distributed training and hyperparameter optimization workflows