Senior Machine Learning Engineer (Reinforcement Learning/World Model)

Path Robotics is hiring Senior Machine Learning Engineers to develop neural world models and reinforcement learning systems for robotic welding. The roles focus on enabling robots to predict and optimize manufacturing processes, validate models and policies, and translate research into reliable production systems.

Responsibilities:

• Build action-conditioned world models that predict how the welding process evolves under changes to robot motion and process parameters
• Model relationships among inputs, system state, physical dynamics, and resulting weld quality
• Develop multimodal models using data such as video, 3D scans, thermal measurements, electrical signals, robot state, and process parameters
• Explore latent dynamics, video prediction, generative modeling, and spatiotemporal representations
• Improve long-horizon rollout accuracy, physical plausibility, temporal consistency, and computational efficiency
• Quantify model uncertainty and identify conditions under which predictions are unreliable
• Validate learned predictions against real-world welding data
• Integrate the model into RL, planning, process-optimization, evaluation, and synthetic-data workflows
• Prevent downstream optimization systems from exploiting inaccuracies in the learned model
• Translate promising research into scalable training and inference systems
• Develop reinforcement learning approaches for optimizing welding decisions and process outcomes
• Define state, observation, action, and reward representations based on measurable manufacturing objectives
• Train and evaluate policies using learned world models, traditional simulation, offline datasets, and controlled real-world experiments
• Develop offline, model-based, or constrained RL methods suitable for limited and expensive physical interaction
• Optimize across competing objectives such as weld quality, cycle time, reliability, energy use, and equipment constraints
• Design methods that account for uncertainty, distribution shift, delayed outcomes, and sparse or imperfect reward signals
• Diagnose reward exploitation, unsafe behavior, policy instability, and model exploitation
• Establish reliable offline and real-world policy evaluation methods
• Partner with controls, welding, robotics, world-model, data, and ML infrastructure engineers
• Translate research prototypes into dependable training, evaluation, and deployment systems

Requirements:

• Master's or PhD in Computer Science, Robotics, Machine Learning, or related field, or equivalent practical experience
• Experience developing and deploying reinforcement learning algorithms on real-world systems
• Proficiency in Python and deep learning frameworks such as PyTorch or TensorFlow
• Experience with simulation environments (e.g., MuJoCo, Isaac Gym)
• Solid understanding of probability, statistics, and optimization
• Experience with training and deploying ML models in production systems