Girder AI

Machine Learning Engineer — Robotics (Hugging Face + Isaac Sim)

Girder AI

San Francisco•On-site•3-6 Yrs•4 days ago

We're looking for a Machine Learning Engineer to train, fine-tune, and deploy learned models for our robotics stack.

Fine-tune and adapt foundation models — vision-language models, vision-language-action (VLA) policies, and perception models — using the Hugging Face ecosystem (Transformers, PEFT/LoRA, Accelerate, Datasets)
Build training pipelines for imitation learning and robot policy learning, including data collection, curation, and versioning (e.g., LeRobot-style datasets)
Generate and leverage synthetic training data from Isaac Sim, including domain randomization for sim-to-real transfer
Evaluate policies and perception models in Isaac Sim / Isaac Lab environments: define metrics, build eval harnesses, and run closed-loop rollouts
Optimize models for deployment on edge GPUs (quantization, distillation, TensorRT/ONNX export)
Track, analyze, and iterate on experiments with tools like Weights & Biases or MLflow
Collaborate with simulation and robotics engineers to close the loop between data generation, training, sim evaluation, and hardware deployment
Stay current with the physical AI landscape (VLAs, world models, robot foundation models) and rapidly prototype promising approaches

3+ years of experience in machine learning engineering, with at least some of that applied to robotics, autonomy, or embodied AI
Deep, hands-on experience with the Hugging Face ecosystem: Transformers, Datasets, PEFT, Accelerate, and the Hub workflow (pushing/pulling models, datasets, spaces)
Strong PyTorch skills and solid understanding of transformer architectures, fine-tuning techniques, and training dynamics
Working experience with NVIDIA Isaac Sim (and ideally Isaac Lab) — comfortable setting up scenes, running policies in sim, and using sim outputs for training or evaluation
Strong Python engineering fundamentals: clean code, testing, reproducible environments
Comfort with Linux, GPUs, and distributed or multi-GPU training
Sim-to-real transfer experience: domain randomization, system identification, or real-robot deployment
Familiarity with ROS 2 and real-time inference on embedded platforms (Jetson, etc.)
Contributions to open-source ML or robotics projects

Browse Jobs