We are building the next generation of physically reliable, safety-critical humanoid robots for real manufacturing environments. You will lead research at the intersection of world models, VLAs, and Deep RL, developing algorithms and control systems that enable humanoids to perform robust manipulation, locomotion, and long-horizon task execution in dynamic, unstructured settings. Your work will define our autonomy roadmap and directly impact real-world deployment.
What You Will Do:
Core Research
- Develop next-gen world models, VLA policies, and Deep RL algorithms for dynamics understanding, uncertainty reasoning, and long-horizon planning.
- Design physics-consistent generative models (e.g., Genie, V-JEPA, latent dynamics) for predictive control and sim-to-real transfer.
- Build visuomotor policies for industrial manipulation and locomotion.
- Improve robustness, fault tolerance, distribution-shift resilience, and safety across behaviors.
- Apply hybrid approaches (model-based RL, diffusion policies, moment matching, MPC hybrids) to stabilize control.
Simulation & Deployment
- Use Isaac Sim, MuJoCo, and custom simulators to generate training data and evaluate policies.
- Build physics-aware pipelines integrating contacts, proprioception, force/torque feedback, and language-conditioned planning.
- Deploy policies on real humanoids (Unitree-class or custom), debugging failure modes and enhancing reliability and recovery behaviors.
What We’re Looking For
- PhD or Master's with equivalent experience in ML, Robotics, CS, Physics, or related fields.
- Deep expertise in:
World Models (V-JEPA, Genie, latent dynamics)
VLA models (RT-X, Pi-0.6)
Deep RL (actor-critic, model-based RL, large-scale RL)
Generative modeling (diffusion, autoregressive, latent video models)
Hands-on deployment of policies on physical robots.
Strong engineering skills in Python + PyTorch/JAX and scalable training systems.
Experience with humanoids, multi-contact control, whole-body planning, or balance control.
Experience with distributed GPU training and sim-to-real pipelines.
Publications at NeurIPS/ICLR/RSS/CoRL/CVPR.
Familiarity with ROS2, unitree-sdk, LCM, or real-time control loops.
Why Join Us
- Collaborate with top researchers and professors across AI, physics, and robotics.
- Impact real deployments in industrial facilities.
- Competitive compensation, equity, and large-scale compute.
