About The Role

The AI innovation team at Agility works on building and deploying next-generation robot foundation models and end-to-end policies on humanoid robots. Your goal will be to develop and test cutting-edge methods for imitation learning and reinforcement learning on humanoid robots, in order to establish the techniques necessary for humanoid robots to perform different real-world tasks. You will work on a team, running experiments on humanoid robots, and will research and implement methods which can be transferred into production.

About the Work

• Design, train, and deploy robust policies for locomotion, manipulation, and dynamic interactions with the environment.
• Develop core reinforcement learning infrastructure, including scalable training pipelines and evaluation frameworks.
• Design and implement new simulation environments and tasks to support training and deployment of control policies.
• Develop, design, and test imitation learning methods
• Collaborate with Robotics Software and AI engineering teams to develop policies which can be transferred to production

About You

• 3+ years of experience developing and deploying learning-from-demonstration
• Strong programming skills in Python, with proficiency in deep learning frameworks such as PyTorch.
• Experience with modern learning-from-demonstration tools like DiffusionPolicy
• Experience with robot data collection, training, and testing on hardware to perform manipulation tasks.
• Ability to work collaboratively in a fast-paced environment to deliver safe, high-quality software
• MS in Robotics, Computer Science, or a related field.

Bonus Points

• PhD in Robotics, Computer Science, or a related field.
• Publications in top ML or robotics conferences (e.g. NeurIPS, ICML, CoRL, RSS, ICRA).
• Familiarity with robot simulation environments (e.g. Mujoco, Isaac Sim) and sim-to-real transfer techniques.
• Experience with modern reinforcement learning techniques for locomotion, manipulation, and whole-body control
• Experience with writing performant, high quality software in C++

This a hybrid position based out of one of