AI Engineer - Deep Learning for Robot Dexterous Manipulation

Role Overview We are seeking a Deep Learning Manipulation Engineer to drive the next generation of learning-based dexterous manipulation for real-world humanoid robots. In this role, you will design, train, and deploy advanced models that enable high-DOF, multi-fingered systems to grasp, manipulate, and perform complex, long-horizon tasks in unstructured environments. You’ll work closely with robotics, teleoperation, perception, and control teams to transform cutting-edge research into reliable, real-time performance on hardware. This is a rare opportunity to join at the ground floor and help shape the entire manipulation learning strategy and infrastructure for a rapidly scaling humanoid company. Responsibilities • Design and implement state-of-the-art deep learning models for dexterous manipulation, including grasping, tool-use, in-hand manipulation, and multi-step tasks. • Develop curriculum-based training procedures that scale from simple interactions to complex, long-horizon behaviours. • Integrate tactile sensing, proprioception, and multimodal inputs into end-to-end closed-loop learning pipelines. • Work with teleoperation and data teams to define data collection, storage, and versioning strategies for large-scale manipulation datasets. • Leverage and extend current SOTA manipulation architectures (e.g., VLA models, diffusion policies, foundation models) to tackle emerging challenges. • Deploy learned manipulation models onto humanoid hardware, ensuring real-time performance, safety, and seamless integration with sensors and control stacks. • Build, optimise, and maintain the manipulation ML training, simulation, and evaluation pipeline. • Develop rigorous evaluation frameworks for both sim and real-world performance, including stress-testing for robustness and generalisation. • Stay on top of the latest research in deep learning, dexterous manipulation, and robotics, and bring new ideas into production. • Contribute to the growth, mentorship, and technical direction of the machine learning and autonomy teams. Qualifications • Master’s or PhD in Robotics, Computer Science, or related technical field. • 3+ years applying deep learning to robotic manipulation tasks. • Strong expertise in behaviour cloning, diffusion models, VLA models, foundation models, or similar SOTA approaches. • Experience working with large datasets, cloud-based training, and large-scale compute. • Hands-on experience deploying learned models in the real world and understanding the constraints of on-hardware inference. • Proficiency in Python and modern DL frameworks (e.g., PyTorch). • Strong software engineering practices and a first-principles mindset. • Ability to thrive in fast-moving, ambiguous, highly collaborative environments. • Persistence, grit, and a passion for tackling some of the hardest problems in robot manipulation. Preferred Experience • Background in computer vision, perception algorithms, segmentation, detection, or point-cloud processing. • Publications in top ML/robotics venues (e.g., NeurIPS, ICML, CoRL, ICRA, RSS). • Experience deploying robots, collecting large-scale datasets, and training large neural networks for production systems. Benefits • Work on transformative humanoid robotics technologies with real-world impact. • Join an early, fast-moving team where your contributions directly shape product direction and technical foundations. • Collaborate with world-class engineers who value creativity, ambition, and continuous learning. • Access state-of-the-art labs, tools, and prototyping environments. • Competitive compensation, excellent benefits, flexible working environment, and equity opportunities. • A culture built on inclusivity, diverse perspectives, and shared ownership. If you're driven to push the boundaries of real robotic intelligence and want to help build a humanoid system capable of meaningful work in the real world, we’d love to hear from you.