Description
Humanoid is hiring a Research Engineer on its World Models team to build and train action-conditioned generative models for robotics. The role focuses on video, state, action, and language models using diffusion and transformer architectures; action-conditioned video prediction and dynamics; learned-simulator evaluation; synthetic rollouts and counterfactual experience; fidelity metrics; data pipelines; and scaling studies. The engineer will collaborate with pretraining, reinforcement learning, and manipulation teams and validate models against industrial fleet data. The position requires a track record of training large generative models, hands-on experience with modern generative architectures and distributed training, strong Python and PyTorch/JAX skills, and empirical rigor.
