We are building scalable, structured foundation models for robotic embodiment with natural inductive biases. We’re an early-stage startup, backed by top investors and led by researchers and operators from Berkeley, AMLab, and Google. Learn more at https://www.newtheory.ai/.
We’re looking for researchers with a proven track record building geometric deep learning, vision-language-action, and reinforcement learning foundation models. Even better if you have a history of turning novel research into reliable results on real robots. In this role, you’ll design and build state of the art models for efficient, scalable, real world tasks. If you can think outside of the box, pair rigorous experimental methods with strong engineering, and ship low-latency behaviors on physical platforms, you’ll succeed in this role.
We’re looking for researchers with a proven track record building world models that enable long-horizon planning, memory, and scene understanding. In this role, you’ll prototype planning and simulation components, define and test benchmarks for forecasting, and integrate predictions with low-level policies to deliver measurable robot performance. If you can translate geometric representations into robust real-world behavior and iterate quickly from paper to production code, you’ll succeed in this role.
We’re looking for researchers with a proven track record in robotics sensing, controls, calibration, runtime, and reliability. In this role, you’ll bring new robotics platforms online, integrate learned models on diverse hardware, harden for latency/throughput, and validate end-to-end performance. If you can bridge algorithms and systems, debug at the sensor/firmware/OS layers, and make complex stacks stable under real-world constraints, you’ll succeed in this role.
We’re looking for experienced deep learning engineers to build high-reliability data and training pipelines that keep our embodied models improving continuously. In this role, you’ll own ingestion and labeling for multimodal data, run large-scale training/inference (scheduling, checkpoints, metrics, reproducibility), and turn research prototypes into scalable jobs that ship to robotic hardware. If you can treat ML pipelines as production software and still move fast with researchers, you’ll succeed in this role.
<aside> 💭
If you have something to contribute to New Theory and you don’t see the right role here, we would love to explore working together. Fill out the form below and choose “Make My Own Role”.
</aside>