At Boston Dynamics, we are developing the next generation of intelligent robots capable of operating in complex, human-centric environments. A critical component of this effort is enabling our robots to understand and interact with the 3D world around them. We are seeking a talented Research Scientist to join our Atlas Controls team and pioneer new methods at the intersection of 3D perception and reinforcement learning.

As a Robotics Research Scientist , you will be at the heart of solving one of the most challenging problems in robotics: teaching a humanoid robot to perform complex loco-manipulation tasks in unstructured environments. Your work will focus on developing novel reinforcement learning policies that navigate and interact with the world. This is a unique opportunity to translate cutting-edge research into real-world capabilities on one of the world's most advanced robots.

What You'll Do

• Design and implement novel reinforcement learning algorithms that leverage environmental perception to solve complex locomotion and manipulation tasks on real world humanoid robots.
• Leverage high-fidelity simulation environments extensively to develop and validate control policies before deploying them on physical hardware.
• Collaborate closely with controls, perception, and software teams to integrate your policies into the broader robot software stack.
  
  

We're Looking For

• Ph.D. in Robotics, CS, or a related field with a focus on developing and training reinforcement learning policies for legged robot locomotion or manipulation; OR a Master's degree with 3+ years of hands-on professional experience deploying RL policies on legged robots or robotic manipulators.
• Technical understanding of 3D geometry, computer vision, and data structures for representing 3D scenes.
• Extensive experience developing and testing RL agents in simulation environments (e.g., Isaac Sim, MuJoCo).
• Strong proficiency in Python and C++.
• A solid understanding of robotics fundamentals, including kinematics, dynamics, and coordinate frames.
  
  

Nice-to-have

• Demonstrated experience deploying RL policies on physical robotic systems.
• Experience integrating rich perceptual data (e.g., vision, depth) into a control or learning-based policy.
• A passion for building robust and reliable software for real-world robotic systems.
• Publication record in top-tier robotics, machine learning, or computer vision conferences (e.g., CoRL, RSS, ICRA, CVPR, NeurIPS).