Robotics Foundation Model Engineer

Join Physical Intelligence to build the first truly general-purpose brain for the physical world. As a Robotics Foundation Model Engineer, you will develop and scale large-scale multimodal models that enable diverse robotic hardware to perform complex tasks in unstructured environments. Your work will directly impact our core mission of creating universal robot policies that generalize across different embodiments, from industrial arms to mobile manipulators.

- Design and train large-scale foundation models that ingest multimodal data (vision, language, proprioception) to output low-level robot control. - Develop scalable data pipelines to ingest and process heterogeneous robotics datasets from varied sources and embodiments. - Optimize model architecture for real-time inference on edge devices without compromising generalizability. - Carry out rigorous evaluations of model performance across both simulated environments and physical hardware deployments. - Collaborate with the hardware team to define telemetry requirements that improve model training efficiency. - Research and implement state-of-the-art techniques in tokenization for continuous control and cross-embodiment mapping. - Fine-tune foundation models for specific downstream tasks using few-shot or zero-shot learning techniques.

- MS or PhD in CS, Robotics, or related field with a focus on Deep Learning. - Proven track record of training large-scale models (GPT, PaLM, Flamingo) or large-scale Vision-Language-Action (VLA) models. - Deep expertise in PyTorch or JAX, specifically centered on distributed training across massive GPU clusters (deepspeed, FSDP). - Understanding of robotics fundamentals including kinematics, control theory, and simulation-to-real (Sim2Real) transfer. - Experience with imitation learning, reinforcement learning, or diffusion-based policy architectures. - History of contributing to open-source breakthroughs or high-impact publications in the AI/Robotics space. - Ability to work independently in a fully remote, highly collaborative global environment.