Builds and ships multimodal perception models that run reliably on physical robots at OpenMind, which is building the operating system for robotics. Day to day: training/evaluation/inference pipelines, containerized model services, and optimizing foundation models for real-time inference on embedded hardware with PyTorch/JAX, TensorRT/ONNX, ROS2, Docker, and NVIDIA Jetson.
OpenMind is building the operating system for robotics: the models that let robots understand people, context, and norms well enough to work alongside humans safely. Our models run on real robots with real customers, and every deployment produces new data to learn from.
You will be the engineer who turns research into models that run reliably on physical robots. You will build the training, evaluation, and inference systems behind our models. You will work with a continuous stream of real-world embodied data, including egocentric video, lidar, audio, state-action logs, and the robot's own reasoning traces, collected from humanoids and quadrupeds operating in public spaces.
WHAT YOU WILL DO
Build multimodal perception models that make robots socially aware, such as active speaker detection, person tracking across camera and lidar, and engagement estimation
Build and scale the pipelines that turn multi-hour, full-stack robot logs into curated, labeled datasets for training, evaluation, and external data partnerships
Ship models as containerized services in our OM1 deployment stack, and own their latency, reliability, and regressions in the field
Work closely with our research lab, robotics engineers, and deployment teams to move models from prototype to production on physical robots within weeks
Integrate, fine-tune, and optimize foundation models for real-time perception and reasoning on embedded compute
WHAT WE LOOK FOR
MS or PhD in machine learning, robotics, computer science, or a related field
1+ years of industry or applied research experience training and shipping machine learning models
Strong engineering fundamentals in languages such as Python, Go, and/or C++, and fluency in PyTorch or JAX
Hands-on experience with multimodal models, perception, or reinforcement and imitation learning
Experience running models on edge hardware under real latency and memory constraints, such as the NVIDIA Jetson software stack
Experience with TensorRT, ONNX, or other inference optimization toolchains
Experience with ROS2 and Docker-based deployment on robots
Experience building datasets or benchmarks used outside your own team
Publications at CoRL, RSS, ICRA, NeurIPS, ICML, or ICLR
A bias toward models that work in the real world, not only on benchmarks