Robot Learning Engineer | Humanoid

Role: Robot Learning Engineer Company: Humanoid Location: London, UK (King’s Cross) – on-site, five days a week Job Type: Full-time About Humanoid Humanoid is building the world’s most capable, commercially scalable and safe humanoid robots. Its platform, HMND-01, is already deployed in real industrial environments, which makes Humanoid the most serious humanoid robotics company in Europe. • 300+ people, 34+ nationalities, flat structure: your decisions ship directly into real robots. • Offices in London (HQ, King’s Cross), Boston, Vancouver and San Diego. • Backed by serious funding, growing fast and deliberately. • Values: Speed, Ambition, Passion, Trust. About the role The Manipulation team sits at the center of everything Humanoid does. In their own words, they are the bottleneck: the work is intense, the impact is immediate, and from day one you get access to prototype hardware worth hundreds of thousands of pounds. The team works on whole-body locomotion and manipulation, using imitation learning and RL on real robots. This is a senior or staff individual contributor role with no management track, and the team is hiring for several openings. It is not a research role: the goal is policies that work on the robot, and results are shared as blog posts, not papers. What you are expected to do • Post-train manipulation policies with behaviour cloning and RL, and own the full loop from raw data to deployment on a real robot. • Work with the simulation team to set up RL training on a digital twin. Iterate on reward design and simulation quality to get reliable sim-to-real transfer. • Partner with the Data Collection org to drive collection: define what good data looks like, ensure diversity and coverage, iterate on operator instructions. • Design data preprocessing strategies that improve the quality of collected data. • Expand observation and action spaces to support new capabilities, and work with the Teleoperations team to expose them to robot operators. • Work with the Controls teams to improve motion smoothness and the teleoperation experience. • Feed what you learn about current hardware limits back into future hardware design cycles. • Work closely with the pre-training team to take advances from research and make them work reliably in real-world industrial demos. What you need to succeed • 3+ years working on robots in industry or research, with shipped work to show for it: a track record, not just theory. • Hands-on experience with modern manipulation methods, ideally close to production deployment. • Solid understanding of modern teleoperation and low-level robot control stacks. • Experience with neural network post-training: behaviour cloning and RL applied to physical systems. • Familiarity with deep learning infrastructure: streaming datasets, checkpointing, distributed training, PyTorch or JAX. You can profile and debug numerics. • Strong software engineering and coding fundamentals: maintainable research code, clear experiment docs, crisp communication of trade-offs. The process includes a LeetCode-style coding round, so come prepared. • You are driven by getting things to work on the robot, not by writing the perfect algorithm in simulation. When something breaks, your instinct is to debug it, not publish it. • You are proactive: you can take an open problem and drive it without waiting for a spec. • Working English. Slightly below the experience bar but current on modern manipulation methods? Apply anyway. Strong PhD students close to finishing and postdocs are welcome too. Nice to have • Experience training VLA models for manipulation (autoregressive, diffusion or flow-matching based); familiarity with OpenVLA and Physical Intelligence (π) models. • Experience applying RL to robotic probl