Remote job
Robotics ML Research Engineer
Job details
About this role
Role overview Develop a practical learning loop that uses robot recordings to produce and measure policy improvements. This applied research and engineering role covers dataset design, model training and evaluation, in close collaboration with robotics and platform engineers; its output is a reproducible process that can be run and tested.
Responsibilities - Define data collection tasks, coverage and instrumentation so recordings capture useful training and evaluation examples, including failures and recovery. - Build dataset construction and quality-control pipelines with clear episode boundaries, metadata, provenance and reproducible versions. - Train baseline and improved models from multimodal trajectories, investigating data quality, representations and learning objectives. - Design controlled evaluations that reveal generalization limits and regressions and connect data changes to outcomes. - Deploy and test policies with robotics leads, distinguishing model problems from sensing, control or integration issues. - Use real training workloads to improve data retrieval, export and processing; maintain tested code and communicate findings and next experiments.
Requirements - Deep robotics or machine learning experience, ideally at PhD level or equivalent practical research experience, including hands-on work with real robots. - Experience training models on multimodal data and diagnosing failures through careful experimentation. - Strong Python skills and experience with modern deep-learning tools, plus the ability to write production-quality research and data-processing code. - Rigorous evaluation and reproducibility practices, clear English communication, and comfort working across research, hardware and platform teams.
Nice to have Experience in manipulation learning, world models, imitation or reinforcement learning, ROS 2, sensor synchronization, data-centric machine learning or automated robot evaluation.
Benefits and work setup The role offers ownership across the learning loop and the opportunity to shape how data supports improvements in robot behavior, with competitive salary and equity.