Job Description

About This Role

RoboSpec is looking for an Embodied AI Engineer to build, train, and deploy AI models that drive real robotic manipulation. You will own the path from model to motion. That covers developing and fine-tuning VLA, VLM, and imitation learning policies, and making them run reliably on physical manipulator arms such as Franka and UR. You will work on RoboSpec's core embodied AI stack. The goal is robotic intelligence that is capable, efficient, robust, and deployable in the real world.

Key Responsibilities

Model Development & Training: Develop, train, and fine-tune foundation models for robotic manipulation, including VLA, VLM, action models, and multimodal models. Design and improve model architectures, training methods, and data pipelines.
Robot Learning: Apply imitation learning (e.g., Diffusion Policy, ACT), reinforcement learning, or behavior cloning to real-world manipulation tasks. Explore and prototype state-of-the-art methods in embodied AI.
Real-Arm Deployment & Control: Deploy trained models on physical cobots such as Franka and UR. Build and maintain ROS 2-based control and integration components in Python and C++.
System Integration: Integrate perception (cameras, sensors), manipulation primitives, learned policies, and control modules into a unified, reliable robotic system.
Performance & Robustness: Optimize model inference efficiency, generalization, and robustness. Debug and profile control-loop performance, address real-world latency, and close sim-to-real gaps on hardware.
Simulation: Use simulation tools such as Isaac Sim or Gazebo for policy training, data generation, and behavioral testing before real-world deployment.

Requirements

Master's degree or PhD in Computer Science, AI, Robotics, Machine Learning, Computer Vision, or a related field.
Strong background in deep learning, with hands-on PyTorch experience and familiarity with modern architectures such as Transformers and diffusion models.
Experience in at least one of the following: multimodal learning, foundation models, or robot learning.
Hands-on experience programming and controlling real robotic arms (Franka, UR, or similar).
Solid experience with ROS 2 and robotics fundamentals, including kinematics, dynamics, trajectories, control, and sensing.
Strong programming skills in Python. C++ is a plus.
Ability to independently design, implement, and validate new models and systems end to end.

Preferred Experience

Vision-Language-Action (VLA) models or multimodal foundation models
Diffusion models, flow matching, or world models
Sim-to-real transfer, including domain randomization or adaptation
Distributed training and GPU optimization
End-to-end delivery of real-world robotics applications such as advanced picking or assembly
PhD, or publications at leading AI, vision, or robotics conferences (e.g., NeurIPS, ICML, CVPR, CoRL, RSS, ICRA)