Research Scientist

ppDo you believe the path to general-purpose physical AI runs through noisy, real-world factory deployments?br/Are you excited by the challenge of turning the classical robotic stacks into the foundational training data for physical AI?br/Do you want to bridge the gap between world-class ML research and industrial-scale robotic execution? /p pIf your answers are yes, we should talk. /p pAt Nomagic, we are executing a humble pivot for general-purpose physical AI. We believe that physical AI is fundamentally a knowledge transfer problem - we are leveraging the "internet data" of robotics - massive deployment logs from real systems operating in production environments - to bootstrap our efforts. We are looking for Research Scientists who will help us to build, train, and deploy foundational models that bring our fleet from a classical control stack to generalized AI mastery. /p h3Offer essentials /h3 ul liPlay with real robots, solving real problems, every day. /li liRelocation package. /li liFlexible working hours. /li liEnglish-speaking environment. /li /ul h3Why we love this job /h3 ul lipWe combine world-class research with top-notch engineering and apply it to solve real problems. /p /li lipMuch of this data already exists. We have robots in production at scale. We aren't waiting for datasets to be collected; the byproduct of our machines doing useful work is being created right now. /p /li lipWe measure what matters. We test our code in unit tests, simulations, and directly on real robots. Grounding our models in deployment allows us to truly measure performance, not just offline metrics. /p /li lipHigh leverage, high impact. We’re still a highly focused team. If your architectures and training curricula improve our agents, you directly change the economics of the company. /p /li lipWorld-class peers. Our team has built Google Warsaw, unicorn startups, led research in DeepMind, tested rocket engines, and worked at top companies like Nvidia and ByteDance. Now, we are shaping the reality of Physical AI together. /p /li lipWe are building the bridge. We aren't a new startup looking for an application; we are an established player bootstrapping physical AI. We believe this will be the first true proof-of-concept for scaled physical AI. /p /li /ul h3What you will do /h3 ul lipYour focus will be defined by the intersection of ML research, robotics, and large-scale multimodal model training. Expect challenges across two main pillars, with the opportunity to specialize in Pretraining or Post-training: /p /li lipFoundation Models Pretraining /p ul lipDesign the Base Intelligence: Define model architectures (Transformer- and Diffusion-based), objectives, and training curricula across multimodal robotic data, turning raw deployment logs into generalizable capabilities. /p /li lipMaster the Data: Develop scalable data mixtures and sampling strategies utilizing our massive offline repositories of vision, action, and state data. /p /li lipPush the Frontier: Run rigorous ablations to understand scaling laws, data quality effects, optimization dynamics, and large-model failure modes. /p /li lipScale with Engineering: Collaborate closely with ML Infra to push cluster utilization and throughput, ensuring our algorithmic ideas translate to efficient distributed training. /p /li /ul /li lipAdaptation, Post-Training Real-World Evaluation /p ul lipDrive Downstream Adaptation: Explore fine-tuning recipes to make general models – our own as well as our partner’s models – useful, controllable, and safe in the real world using imitation and reinforcement learning, distillation, and curriculum learning. /p /li lipImprove Physical Robustness: Develop cutting-edge methods for improving real-world reliability, handling out-of-distribution edge cases, and steering robot behavior in mature factory environments. /p /li lipBuild Benchmarks: Design evaluation frameworks and lightweight physical setups that measure actual robot performance and failure modes far beyond the limits of simulation. /p /li lipClose the Physical Loop: Analyze real-world evaluation results to guide the overarching research direction, seamlessly bridging the gap between foundation model outputs and physical-world outcomes. /p /li /ul /li /ul h3What skills we’d like you to have /h3 ul lipExperience: Deep research and practical experience at the intersection of machine learning, systems engineering, and physical robotics. /p /li lipProven Track Record: Experience designing, training, and fine-tuning large-scale deep learning architectures (VLMs, VLAs, RL, RLHF, Imitation Learning), ideally with policies deployed and validated on real hardware. /p /li lipEngineering Excellence: Strong deep learning framework fundamentals (PyTorch/JAX). You are comfortable debugging at every layer of the stack and care about empirical rigor as much as raw iteration speed. /p /li lipRobotics Intuition: Comfort working hands-on with hardware. You understand the robotics full stack (perception, controls, state estimation) and care deeply about evaluation and failure analysis when software meets the physical world. /p /li lipPragmatic Research Mindset: You possess the ability to move seamlessly between theoretical design and physical implementation. You prefer execution, rapid iteration loops, and real-world robustness over academic purity. /p /li /ul h3Application process /h3 ul liA phone screen with the hiring manager to discuss your background and our technical direction. /li liA half‑day of on‑sites (cultural fit deep‑divo technical interviews). /li liA final decision made within 2‑3 days after the on‑site interview. /li liImportant: Expect detailed, honest feedback after completing the process, regardless of our decision. /li /ul /p #J-18808-Ljbffr