Robotics AI Research Engineer
About the role
Join our Humanoid Robotics R&D team to develop the intelligence behind next-generation humanoid robots. You will research, train, fine-tune, evaluate, and deploy Vision-Language Models (VLMs) and Vision-Language-Action (VLA) models, connecting foundation-model research with real robotic hardware.
Key responsibilities
Fine-tune and post-train VLM / VLA foundation models using internal humanoid robotics datasets.
Develop SFT, preference-optimization, reinforcement-learning, and imitation-learning pipelines using simulation and real-hardware feedback.
Research hierarchical robotic policies combining high-level reasoning, skill planning, and low-level whole-body control.
Run large-scale distributed training in multi-GPU environments using frameworks such as FSDP / DeepSpeed.
Design robotics data strategies covering teleoperation, motion capture, human video, synthetic simulation, and external datasets.
Build reproducible evaluation pipelines in simulation and on real humanoid hardware.
Drive sim-to-real transfer through domain randomization, system identification, distillation, and policy learning.
Deliver production-ready checkpoints and inference pipelines for edge and GPU environments.
Collaborate with Data Factory and Platform Engineering teams to convert research into deployable robot capabilities.
Mentor engineers and contribute to technical research, patents, or publications where appropriate.
About you
MSc or PhD in Computer Science, Robotics, Machine Learning, Electrical Engineering, or related discipline, or equivalent industrial research experience.
3+ years of hands-on experience training or post-training large neural networks using PyTorch.
Experience with at least one of: Vision-Language Models, Vision-Language-Action models, diffusion / flow-matching policies, or large-scale imitation learning.
Experience with modern LLM/VLM frameworks such as Hugging Face Transformers, TRL, LLaMA-Factory, Axolotl, Unsloth, or equivalent.
Experience with distributed training frameworks such as FSDP, DeepSpeed, or Accelerate.
Strong foundations in robotics, including kinematics/dynamics, control, imitation learning, and reinforcement learning.
Experience with robotics simulation such as Isaac Sim, MuJoCo, PyBullet, or Genesis.
Strong Python engineering and modern software/MLOps practices.
Proven ability to move trained models beyond research experiments into real systems.
Strong English communication skills.