AI Research Engineer

Description

The team builds machine learning systems that let robots learn dexterous manipulation from human demonstrations and multimodal sensor data.

Work spans data processing, model training, evaluation, and deployment on real robots, with experiments aimed at finding what helps robots learn better.

The research covers vision, motion, biomechanical signals, and robot interaction data, applied to imitation learning, vision-language-action models, video models, and world models.

The company is a small research and engineering startup working on how robots can learn skills from human biomechanics. The founders have AI research backgrounds at MIT and NVIDIA, and deployment experience in autonomous vehicles and robotics at Pony AI and Dexterity.

You will work directly with the founders, own problems from start to finish, and test ideas on real robotic systems.

Prior robotics experience is not required.

Responsibilities

• Build and refine training, fine-tuning, and evaluation pipelines for models that use human and robot data.
• Train and adapt models for manipulation, action prediction, and learning from demonstrations.
• Find ways to learn representations from vision, motion, and biomechanical signals.
• Test imitation learning, reinforcement learning, and foundation models on robotic tasks.
• Plan experiments and ablations to see how models behave, where they fall short, and how much they improve.
• Create scalable data processing, training, and evaluation infrastructure for fast iteration.
• Work with robotics engineers to deploy models on physical robots, study failures, and improve results from real-world feedback.
• Read and reproduce relevant research, and adapt useful methods to new problems.

Requirements

• Strong Python skills and experience with PyTorch, JAX, or similar machine learning frameworks.
• Solid grasp of deep learning, model training, optimization, and evaluation.
• Experience building and evaluating machine learning systems, not only calling pretrained models through APIs.
• Ability to design experiments independently, debug training pipelines, and interpret results.
• Experience in at least one area such as computer vision, multimodal learning, generative models, representation learning, reinforcement learning, or imitation learning.
• 3+ years of experience.
• Experience with large-scale training, distributed computing, or ML infrastructure is a plus.
• Experience with robotics, VLA models, video models, or world models is a plus, but not required.

Conditions

Full-time position.

On-site in San Francisco.

Visa sponsorship is available.

Equal-opportunity hiring applies.