Robotics AI Engineer – Embodied AI

About the Role

We are building a new robotics team at AXONS focused entirely on the application layer — customization, training, and deployment of service robots and next-generation humanoids. We don't manufacture hardware; we make robots work intelligently in real environments, partnering with world-class hardware providers to bring robots into food processing facilities, retail stores, and logistics operations across the CP Group ecosystem. We are looking for a Robotics AI Engineer who will lead our virtual training pipeline and help shape our approach to data-driven policy learning. In the near term, you will build high-fidelity digital twins of our production environments and train robot policies using Reinforcement Learning and Imitation Learning. Over the next 12–18 months, as the team matures and our real-world data assets grow, you will help us evolve toward a modern framework driven by Egocentric Human Data (EHD) and Vision-Language-Action (VLA) foundation models. This role is for someone who is deeply grounded in simulation and RL today, and genuinely excited about where embodied AI is heading.

Responsibilities

• Phase 1 — Simulation & Reinforcement Learning (Primary Focus): Build, calibrate, and maintain high-fidelity 3D simulation environments representing actual deployment scenarios using NVIDIA Isaac Sim, MuJoCo, Drake or PyBullet; design and optimize RL frameworks for manipulation and locomotion policies, focusing on structured pick-and-place tasks with bimanual robot arms; develop sim-to-real transfer and domain randomization strategies to ensure reliable policy transfer to physical hardware; handle non-rigid object manipulation for compliant and variable objects like organic materials and flexible packaging.
• Phase 2 — Data Flywheel & Evolving Paradigm (12–18 Month Horizon): Help transition from pure simulation-based training toward pipelines seeded by real human demonstration data, build behavioral cloning pipelines to warm-start RL, research and integrate Vision-Language-Action architectures and egocentric video datasets, and evolve simulation environments into evaluation sandboxes as foundation models mature.
• Evaluation & Infrastructure: Develop evaluation infrastructure and tooling to measure policy performance and support continuous improvement as real-world data is ingested.

Qualifications

• Education: Master's or Ph.D. in Robotics, CS, or related field; or Bachelor's with equivalent industry experience.
• Simulation: Hands-on experience with NVIDIA Isaac Sim, MuJoCo, or PyBullet, including building environments from scratch.
• ML & Deep Learning: Proficiency in PyTorch / TensorFlow, deep RL libraries, and modern architectures like Transformers and Diffusion Policies.
• Robotics Fundamentals: Command of kinematics, transformations, spatial geometry, and object affordance modeling.
• Programming: Advanced Python and working proficiency in C++; CUDA experience is a plus.