Were looking for Research Scientists and Research Engineers to advance the reasoning and planning capabilities of our foundation world models - enabling robots to decompose goals, plan multi-step actions, and handle long-horizon tasks in complex, unstructured environments. We hire across levels - from senior to staff.
What Youll Do- Research and develop methods for multi-step reasoning and planning grounded in embodied world models
- Design architectures and training strategies that improve compositional generalization and long-horizon prediction
- Explore chain-of-thought reasoning, process reward models, and test-time search in the context of robotic control
- Build evaluation benchmarks for reasoning and planning capabilities applied to physical tasks
- Investigate how world model rollouts can enable planning and decision-making at inference time
- Collaborate with pre-training and post-training teams to integrate reasoning capabilities into the full model pipeline
- Publish and present work at top-tier venues (especially valued for RS track)
What Were Looking For- Strong background in reasoning, planning, or search with large models
- Deep understanding of sequence modeling, transformer architectures, and generative models
- Experience with test-time compute methods (beam search, MCTS, self-consistency, verifiers, etc.)
- Strong research taste and ability to identify high-leverage directions
- Fluency with PyTorch or JAX and ability to implement and iterate on research ideas end-to-end
- Staff-level candidates are expected to define technical direction and drive research strategy independently; senior/MTS candidates execute complex projects with strong fundamentals and growing scope
Nice to Have (But Not Required)- PhD in ML, Robotics, or a closely related field
- Publication record at NeurIPS, ICML, ICLR, CoRL, or related venues
- Prior work on reasoning in LLMs (chain-of-thought, process reward models, search-based methods)
- Experience with model-based planning or hierarchical reinforcement learning
- Familiarity with long-horizon prediction, video generation, or world model rollouts
- Experience with embodied AI or robotic planning problems
Why This Role- Tackle one of the hardest open problems in embodied AI: enabling robots to reason about what to do next
- Research that directly translates to robot behavior in complex, real-world scenarios
- High research freedom grounded in real task performance
- Tight collaboration with pre-training, post-training, and robotics teams
