Robotics Engineer Dual-Arm Manipulation & Learning

Robotics Engineer - Dual-Arm Manipulation & Learning
Global Automation Research

About the Project:
The Global Automation Research team at OEM is developing a dual-arm robotic system capable of picking and placing rigid automotive parts from complex dunnage while simultaneously handling the dunnage itself. This system targets one of the hardest open problems in factory automation: replacing manual kitting and sequencing tasks that single-arm robots simply cannot perform reliably.

The work spans the full stack - from hardware integration and teleoperation-based data collection to simulation-based RL policy training and Sim-to-Real transfer. You'll be hands-on with real robots and real factory parts, with a clear roadmap toward deployment at OEM manufacturing sites.

What You'll Do:

Teleoperation Framework: Develop and operate a dual-arm teleoperation system to collect high-quality human demonstration data for complex manipulation tasks (e.g., part extraction from tight dunnage, dunnage manipulation, regrasp operations).

RL Policy Training: Design, train, and evaluate reinforcement learning policies in NVIDIA Isaac Sim, targeting robust dual-arm behaviors including coordinated pick-and-place, dunnage handling, and bimanual alignment.

Sim-to-Real Transfer: Bridge the gap between simulation and the physical robot setup by fine-tuning trained policies and validating performance on real hardware.

Simulation Asset Development: Build and maintain physics-accurate simulation assets - from CAD to USD - to support training and testing of manipulation policies.

Perception Integration: Work with camera systems (Intel RealSense, ZED) and 3D vision pipelines to enable part detection and scene understanding.

System Integration: Integrate the learning stack with ROS 2 / MoveIt 2 for motion planning and execution on the physical dual-arm platform.

Required:

• Hands-on experience training reinforcement learning (RL) or deep RL (DRL) policies for robotic manipulation tasks
• Proficiency with NVIDIA Isaac Sim (or Isaac Lab) for simulation and policy training
• Experience with robot teleoperation systems - hardware setup, data collection pipelines, and demonstration quality assessment
• Strong Python skills; familiarity with ROS 2
• Background in Sim-to-Real transfer for manipulation or locomotion tasks

Nice to Have:

• Experience with imitation learning or learning from human demonstration (e.g., ACT, Diffusion Policy)
• 3D asset creation pipeline: CAD ? USD / URDF ? sim-ready assets (Blender, Omniverse tools)
• Familiarity with 3D vision and scene reconstruction (depth cameras, point clouds, NeRF / Gaussian Splatting)
• Experience with foundation or grasp models (e.g., AnyGrasp, DexGraspNet, or VLA models)
• Edge compute and model optimization experience (quantization, deployment on GPUs)
• Embedded robotics experience (real-time control, force/torque sensing)

Who You'll Work With:
You'll collaborate closely with the OEM Global Automation Research team and our university research partners (University of Michigan), contributing to both physical demos and publishable research. The work environment is fast-moving, hands-on, and research-grade - expect to go from simulation to real hardware within the same sprint.