AI Robotics Developer (World Models, VLA, RL)

Summary We are looking for a highly skilled AI Robotics Developer to build scalable simulation pipelines and train state-of-the-art robotic policies. If you excel at turning physics simulations into rich training grounds and leveraging that data for advanced Reinforcement Learning, World Models, and Vision-Language-Action (VLA) architectures, we want you on our team. What You'll Be Doing - Simulation & Environment Design: Build and maintain highly diverse, physics-accurate simulation environments using MuJoCo, NVIDIA Isaac Sim, or ManiSkill to ensure robust policy generalization. - Synthetic Data Generation: Architect pipelines to generate large-scale, high-quality synthetic "human-like" demonstration and visual data entirely within simulation. - Reinforcement Learning: Utilize generated simulation data to train robust control policies using PPO, GRPO, or advanced model-based RL approaches. - World Models & VLA Development: Train Vision-Language-Action (VLA) models for generalized robotic control. Design and implement predictive world models to enable efficient planning, reasoning, and representation learning in complex environments. - Fine-Tuning & Alignment: Apply Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO) to align robotic behaviors, refine VLA outputs, and improve policy reliability. What We’re Looking For - Deep Simulation Expertise: Hands-on experience with MuJoCo, Isaac Sim, or similar physics engines. You know how to randomize domains and build environments that prevent policies from overfitting. - Policy & Model Training Experience: Strong background in applying RL, training World Models (e.g., Dreamer or JEPA architectures), and developing VLA policies for robotic control tasks. - Modern Architecture Knowledge: Deep familiarity with Transformer-based architectures, large-scale representation learning, and multi-modal integration for robotics. - PyTorch & GPU Proficiency: Comfort building complex training loops and scaling them across GPU clusters. Nice to Have - Experience bridging the "Sim-to-Real" gap and deploying policies trained in simulation onto physical hardware. - Contributions to open-source RL or robotics simulation frameworks (e.g., MolmoAct, ManiSkill). Screening Question (Required) Please answer the following question in your application. Proposals without this answer will not be considered: "What are the top 3 most difficult technical problems you have solved related to simulation or policy training? Describe each in 5 sentences (15 total)." Engagement Details - Type: Contract (with potential for long-term engagement) - Availability: Immediate start preferred If you are passionate about pushing the boundaries of embodied AI through simulation and advanced training methodologies, apply below! Strong candidates will be shortlisted for a personal interview.