Job Details:
Job Description:
Scaling foundation models to the physical world is one of the hardest challenges in AI today. This role moves beyond simulation, which puts intelligence directly onto real hardware. You'll build the end-to-end stack for robotic intelligence as part of the Intel's Open Edge Platform and OpenVINO toolkit. The work is entirely open-source under the Apache 2.0 license to create the bridge between high-level reasoning and real-time physical action. About the Role This role spans the entire lifecycle of physical AI, from training large-scale Vision-Language-Action (VLA) models to building lightweight runtimes for real-time deployment. It's a position for engineers who work at the intersection of high-level research and low-level performance optimization. You'll ensure that complex policies don't just work in a paper, but run reliably on edge hardware with minimal latency.
What You'll Do
- Implement and fine-tune state-of-the-art Vision-Language-Action policies for robotic manipulation and control.
- Export and optimize VLA models for edge deployment without accuracy loss via export pipelines, graph optimization, and precision calibration to ensure policies run at high frequencies on constrained hardware.
- Develop safety runtimes to manage action clamping, velocity limits, and workspace bounds for reliable real-world operation.
- Build and maintain model-agnostic inference APIs that abstract across different robotic platforms and hardware backends.
- Support deployment across a range of edge accelerators and inference runtimes to ensure broad hardware compatibility.
- Write and publish research papers at top-tier venues to contribute to novel findings in vision, physical AI, model optimization, and robotic learning.
Qualifications:
What You Bring
- Extensive experience with PyTorch and PyTorch Lightning, including distributed training for large-scale models.
- Hands-on experience working with robotic systems, including industrial robots, EMR (autonomous mobile robots), and humanoid robots. Strong familiarity with robotics hardware and software stacks such as cameras and depth sensors, robot control interfaces, Robotics Operating Systems, and real-world robot deployment and integration.
- Deep understanding of Vision-Language-Action (VLA) architectures and imitation learning techniques applied to embodied agents and robotic systems.
- Technical expertise in model export and optimization, including graph compilation, operator fusion, and precision calibration for efficient edge and on-device inference.
- Familiarity with inference runtime backends such as OpenVINO, ONNX Runtime, and ExecuTorch.
- Ability to write clean, modular Python code and manage complex deployment dependencies across heterogeneous training, simulation, an