Principal Research Scientist

Chewy Chewy · Retail · MA

Principal Research Scientist to join the Physical AI team, focusing on embodied AI for humanoid and high-DOF robotic systems. The role involves leading research in reinforcement learning, imitation learning, vision-language-action models, multimodal perception, whole-body control, manipulation, and sim-to-real adaptation to develop next-generation robotic intelligence for real operating environments. The scientist will invent algorithms, build evaluation frameworks, guide technical direction, and translate research into capabilities that work on real robots, partnering with cross-functional teams and influencing the physical AI stack.

What you'd actually do

  1. Define and lead Chewy’s scientific roadmap for embodied AI and humanoid robot learning, with emphasis on reinforcement learning, imitation learning, VLA-style policy learning, and modern control approaches for real-world robotic systems
  2. Invent, prototype, and validate novel algorithms for manipulation, loco-manipulation, whole-body coordination, multimodal perception, policy adaptation, and task generalization across environments and robot embodiments
  3. Develop learning systems that combine data-driven policies with principled robotics methods such as planning, control, system identification, estimation, and safety-constrained execution
  4. Drive advances in simulation, synthetic data generation, offline and online evaluation, benchmarking, and sim-to-real transfer to accelerate learning and reduce time to deployment
  5. Build and evaluate robot learning pipelines that leverage vision, language, proprioception, force, and touch to enable robust performance on complex physical tasks

Skills

Required

  • PhD in Robotics, Computer Science, Electrical Engineering, Mechanical Engineering, Applied Mathematics, or a related technical field
  • 10+ years of relevant experience in robotics research, embodied AI, machine learning, controls, or a closely related domain
  • Deep expertise in reinforcement learning
  • Deep expertise in imitation learning or behavior cloning
  • Deep expertise in offline RL or policy improvement from logged data
  • Deep expertise in vision-language-action models or embodied foundation models
  • Deep expertise in robot manipulation and loco-manipulation
  • Deep expertise in whole-body control, MPC, or optimization-based control
  • Deep expertise in system identification, state estimation, or sensor fusion
  • Deep expertise in sim-to-real transfer and domain adaptation
  • Demonstrated track record of developing novel methods that improved real robot performance, not just offline benchmarks
  • Strong publication, patent, and/or open-source record in top-tier robotics or machine learning venues
  • Strong software and research engineering skills in Python and C++
  • Experience in modern ML frameworks such as PyTorch and/or JAX
  • Experience with robotics tooling and simulation environments such as ROS2, Isaac Lab, Isaac Sim, MuJoCo, Gazebo, or equivalent platforms
  • Proven ability to operate as a top-level technical leader in ambiguous environments, influence without authority, and communicate effectively with executive stakeholders

Nice to have

  • Experience training and deploying policies on humanoid robots or other dynamically complex platforms such as legged robots or mobile manipulators
  • Experience with dexterous or force-sensitive manipulation, deformable object handling, or contact-rich tasks
  • Experience building or adapting foundation models for robotics, including multimodal representation learning or action modeling
  • Experience combining learned policies with classical controls, planners, or safety layers for production-grade systems
  • Experience working on warehouse, logistics, or industrial robotics applications
  • Experience leading external collaborations with academic labs, research partners, or strategic technical vendors

What the JD emphasized

  • define and advance Chewy’s research roadmap
  • invent new algorithms
  • build evaluation frameworks
  • guide technical direction
  • turn research ideas into capabilities that work on real robots
  • measurable progress on hardware
  • novel methods that improved real robot performance, not just offline benchmarks
  • Strong publication, patent, and/or open-source record

Other signals

  • embodied AI
  • reinforcement learning
  • imitation learning
  • vision-language-action models
  • multimodal perception
  • whole-body control
  • manipulation
  • sim-to-real adaptation
  • humanoid robots