<p>The AI inference co-design team's goal is to take research models and make them run efficiently on our AI-ASIC to power real-time inference for Optimus humanoid robot programs, with applications extending to Autopilot. This unique role lies at the intersection of AI research, compiler development, kernel optimization, math and HW design tailored to robotic autonomy. You will work extensively with AI engineers and come up with novel techniques to quantize models, improve precision and explore non-standard alternate architectures optimized for edge deployment in humanoid robotics. You will be developing optimized micro kernels using a cutting-edge MLIR compiler and solve the performance bottlenecks needed to achieve real-time latency for Optimus tasks like locomotion, manipulation, and decision-making in dynamic environments. You will work closely with the HW team and bring state-of-the-art HW architecture techniques to our next generation HW SoCs designed for versatile robot behaviors.</p> <ul> <li>Research and implement state-of-the-art machine learning techniques to achieve high performance on Optimus edge hardware for robotic perception, control, and autonomy </li><li>Optimize bottlenecks in the inference flow for real-time robot operations, make precision/performance tradeoff decisions, and figure out novel techniques to improve hardware utilization and throughput in dynamic environments </li><li>Implement/improve highly performant micro kernels for Optimus's AI compute units </li><li>Work with AI teams to design robot-friendly neural network architectures optimized for edge deployment in humanoid robotics </li><li>Collect extensive performance benchmarks (latency, throughput, power) across robotic tasks and work with HW teams to shape the next generation of Optimus inference hardware, balancing performance with versatility for diverse robot behaviors </li><li>Experiment with numerical methods and alternative neural networks architectures tailored to robotic locomotion, manipulation, and decision-making </li><li>Collaborate with the compiler infrastructure to enhance programmability and performance for Optimus's onboard AI systems </li><li>Degree in Engineering, Computer Science or equivalent in experience and evidence of exceptional ability </li><li>Proficiency with Python and C++ </li><li>Experience with AI networks, such as CNNs, transformers, and diffusion model architectures, and their performance characteristics </li><li>Understanding of GPU, SIMD, multithreading and/or other accelerators with vectorized instructions </li><li>Exposure to computer architecture and chip architecture/micro-architecture </li><li>Specialized experience in one or more of the following machine learning/deep learning domains: Model compression, hardware aware model optimizations, hardware accelerators architecture, GPU/ASIC architecture, machine learning compilers, high performance computing, performance optimizations, numerics and SW/HW co-design </li></ul>