Robotics Physical AI Expert

Robotics Physical AI Expert

About the Role

Location China Guangdong Sheng Shenzhen Remote vs. Office Hybrid (Remote/Office) Company Siemens Energy (Shenzhen) Co. Ltd. Organization SE CEO Business Unit Company Development Full / Part time Full-time Experience Level Professional / Experienced

A Snapshot of Your Day

This role develops and optimizes embodied AI and robotics systems using fine-tuning, reinforcement learning, and imitation learning. It builds end-to-end training, evaluation, and deployment pipelines for complex robotic tasks. The position integrates perception, planning, and control components to improve robot performance. It involves close collaboration with software, simulation, data, and hardware teams for real-world deployment. The role also explores and adapts state-of-the-art robotics algorithms for practical business applications.

How You’ll Make an Impact

• Conduct secondary development and post-training on embodied AI pre-trained models: multimodal fine-tuning (Supervised Fine-Tuning/preference optimization/Reinforcement Learning, etc.), LoRA/Adapters, prompting & tool invocation, and action/control interface alignment, to improve task success rate, stability, and regression reliability on target use cases.
• Build a training–evaluation–iteration closed loop for complex/long-horizon tasks: dataset construction and versioning, offline replay, simulation validation, on-robot acceptance test scripts and metrics, enabling rapid issue localization and iteration.
• Apply reinforcement learning/imitation learning to manipulation tasks: assembly tolerances, grasp robustness, path/action optimization, recovery policy learning; build/select training frameworks (Offline RL/Online Fine-tuning/Residual RL, etc.) and define sim-to-real strategies and acceptance criteria.
• Participate in end-to-end embodied “brain” capability design: integrating environment perception model calls with VLA and closed-loop policy/planning with feedback; align interfaces and online constraints with the systems team.
• Collaborate with data/systems/simulation/hardware teams to complete real-robot integration and debugging, problem scoping, and performance optimization, ensuring stable operation under noise and communication delays.
• Track the frontier of embodied AI/robotics algorithms (, ACT, Diffusion Policy, world models, etc.), conduct prototyping and engineering adaptation for business scenarios, and distill reusable algorithm components.

What You Bring

• Master’s degree or above in Computer Science/Computer Vision/Automation/Robotics/Control/Artificial Intelligence, or related fields.
• 5+ years of R&D experience in embodied AI/machine learning/multimodal/reinforcement learning, with project experience delivering end-to-end deployment on real robot platforms for manipulation tasks (training/evaluation/deployment/integration).
• Familiar with robotics perception and manipulation fundamentals: pose estimation, calibration, point cloud/depth processing; understand control interfaces and engineering constraints (real-time/safety/tolerances).
• Strong algorithm foundations: multimodal learning, behavior/policy learning, reinforcement learning, model training and evaluation methods; candidates with Online RL/Offline RL training practice are preferred.
• Able to productionize algorithms: data flywheel/closed-loop, observability/regression, inference performance optimization, and integration with system/control pipelines.
• Proficient in Python with C++ engineering development capability; familiar with PyTorch (or TensorFlow); familiar with ROS/ROS2; candidates with simulation experience (Isaac Gym / Isaac Sim / MuJoCo) or sim-to-real real-robot deployment experience are preferred.
• Preferred: representative achievements such as top-tier conference/journal publications, open-source projects, patents, or pilot cases.