Join our team to push the frontier of dexterous manipulation using large-scale reinforcement learning, imitation learning, and generative models. You’ll lead efforts to train general-purpose control policies for high-DOF hands, build robust pipelines for real-world data collection and sim2real transfer, and help shape the future of embodied intelligence.
Requirements: MS or PhD in Robotics, Computer Science, or related field—or equivalent experience
Expertise in reinforcement learning (e.g. PPO, SAC, DAPG), behavior cloning, or diffusion policy learning
Experience with foundation models or representation learning for robotic control (e.g. RT-1/2, BC-Z, VIMA, RT-Trajectory)
Familiarity with contact-rich manipulation tasks (grasping, in-hand manipulation, tool use, etc.)
Strong software engineering skills in Python and C++ (including JAX or PyTorch, Linux environments)
Experience working with MuJoCo, Isaac Gym/Lab, or Brax for high-performance simulation
Knowledge of sim2real techniques (domain randomization, dynamics adaptation, residual policy learning)
Comfortable training policies on high-dimensional action spaces with vision and proprioception
Experience with hardware deployment and data collection on real robot platforms
-
(+) Familiarity with tactile sensing, vision-language policies (VLP), or differentiable simulation
- Competitive salary and meaningful equity
-
Full health, dental, and vision insurance
-
Access to custom-built dexterous robots and large-scale simulation clusters
-
Collaboration with leading researchers in robotics and AI
-
Backed by YC and top-tier investors
-
High-ownership role with the opportunity to lead core initiatives in real-world robot learning
- Design and implement scalable training pipelines for dexterous manipulation using RL, IL, and hybrid techniques
-
Leverage demonstrations and human priors to accelerate learning of contact-rich tasks
-
Integrate tactile, visual, and proprioceptive feedback into policy architectures
-
Develop sim environments for dexterous hands and object interaction using MuJoCo or Isaac
-
Transfer policies from simulation to real hardware using sim2real and adaptation strategies
-
Explore diffusion-based policy learning and generative models for action synthesis and planning
-
Collaborate across AI, hardware, and perception teams to build closed-loop manipulation systems
-
Publish or contribute to cutting-edge research while delivering production-quality control stacks
