Machine Learning Research Scientist, Mechanical Intuition in Multimodal Models

• Design and implement end-to-end modeling pipelines for machine assembly tasks, building from the ground up rather than adapting existing frameworks. • Run systematic experiments to evaluate architectural variants, data collection and curation strategies, and a range of supervised and reinforcement learning techniques for physical manipulation. • Develop and maintain rigorous evaluation protocols to measure policy performance across assembly scenarios, including generalization to novel parts, configurations, and failure modes. • Explore how modern LLMs and agentic systems can be integrated to support physical reasoning and task planning in assembly contexts. • Collaborate with researchers and engineers across TRI and Toyota's broader ecosystem to connect learning-based systems with real hardware and manufacturing workflows. • Contribute to writing and publishing research results in peer-reviewed venues.

• A PhD in a relevant field such as Computer Science, Robotics, Mechanical Engineering, or a related discipline, completed recently (or nearing completion), with some post-PhD or internship work experience. • A demonstrated track record of implementing non-trivial learning systems — not just running baselines, but building pipelines and components from scratch. • Hands-on experience with policy learning, reinforcement learning, or robot learning, with strong intuitions about what makes these approaches succeed or fail in practice. • Proficiency in Python and comfort working across the full stack of a research project, from data processing to model training to evaluation. • Genuine interest in how physical products are designed and manufactured. • Familiarity with large language models, vision-language models, or agentic AI frameworks, particularly in contexts involving structured reasoning or tool use. • Experience with robot manipulation, motion planning, or sim-to-real transfer. • Exposure to manufacturing processes, assembly planning, or CAD/CAM toolchains. • Experience building or contributing to production-level research codebases.