Machine Learning Engineer, Robot Learning, Loco-Manipulation

Build the Path Forward At Path Robotics, we’re building the future of embodied intelligence. Our AI-driven systems enable robots to adapt, learn, and perform in the real world closing the skilled labor gap and transforming industries. We go beyond traditional methods, combining perception, reasoning, and control to deliver field-ready AI that is risk-aware, reliable, and continuously improving through real-world use. Big, hard problems are our everyday work, and our team of intelligent, humble, and driven people make the impossible possible together. We are standing up a new Robot Learning team focused on whole-body loco-manipulation for precision tasks in heavy manufacturing. We are seeking a Machine Learning Engineer to join us as a founding member. You will be among the first ML engineers on a research stack that does not exist anywhere else in the field built around visual reasoning, learned action policies, and reinforcement-learning fine-tuning from real customer data. What You’ll Do Build the team's robot-learning stack from the ground up. This is a founding role; you are designing the training infrastructure, data pipelines, simulation environments, model architectures, and deployment workflows — not inheriting them. Multi-modal perception, scene understanding, and learned action generation work in tight coordination on the stack you help create. Stand up ML infrastructure — training pipelines, experiment tracking, data versioning, reproducible sim-to-real workflows. Train policies across manipulation, locomotion, and the whole-body control coupling between them. On legged platforms performing precision tasks, manipulation and locomotion are not separable — every arm motion shifts the centre of mass; the whole-body controller compensates in real time to maintain accuracy at the tool. Behavioural cloning, diffusion- and flow-matching action generation, reinforcement-learning fine-tuning. Cobots, industrial arms, and mobile platforms. Deploy in stages — through a phased rollout strategy that builds production trust over time. Every real-world execution accumulates training data for continuous improvement. Collaborate daily with mechanical engineers, perception engineers, robotics engineers, and manufacturing domain experts. Within-department rotation across home teams is expected. Who You Are Ph.D. or Master's degree in Robotics, Mechanical Engineering, Electrical Engineering, Computer Science, or a related field — or equivalent experience. 2+ years of hands-on robot learning experience. You have trained policies and deployed them on real robot hardware — not just in simulation. Sim-to-real transfer experience — built simulation environments, implemented domain randomisation, transferred policies to physical robots, debugged where it broke. Implementation experience with diffusion-based or flow-matching action policies for robots, and with action chunking. Reinforcement learning for robotics applied on real hardware — sample-efficient on-robot methods, residual RL on top of pretrained policies, on-policy fine-tuning of foundation policies. Strong programming skills in Python; PyTorch and ML training infrastructure at production level. Practical experience with NVIDIA Isaac Sim / Isaac Lab, MuJoCo, or equivalent. Comfort with physical robots — debugging, iterating, deploying. Strong communication skills, able to convey complex technical concepts to a diverse audience. Strongly Preferred: Edge inference on edge-class hardware (TensorRT, ONNX, FP16 / INT8 quantisation). Real-time on-robot deployment is a core requirement. Visual self-supervised representation learning experience on robot or 3D-vision tasks. Legged-robot or whole-body control experience — locomotion, manipulation on a floating base, or the integration between them on quadrupeds or humanoids. Physics-informed ML — hybrid models where learned components are constrained by known physics. Experience building ML pipelines or infrastructure in a team setting. Why