Research Scientist, World Models - Policy Training and Evaluation
About the position At Toyota Research Institute (TRI), we're on a mission to improve the quality of human life. We're developing new tools and capabilities to amplify the human experience. To lead this transformative shift in mobility, we've built a world-class team in Energy & Materials, Human-Centered AI, Human Interactive Driving, Large Behavioral Models, and Robotics. Within the Human Interactive Driving division, the Extreme Performance Intelligent Control department is working to develop scalable, human-like driving intelligence by learning from expert human drivers. This project focuses on creating a configurable, data-driven world model that serves as a foundation for intelligent, multi-agent reasoning in dynamic driving environments. By tightly integrating advances in perception, world modeling, and model-based reinforcement learning, we aim to overcome the limitations of more compartmentalized, rule-based approaches. The end goal is to enable robust, adaptable, and interpretable driving policies that generalize across tasks, sensor modalities, and public road scenarios-delivering transformative improvements for ADAS, autonomous systems, and simulation-driven software development. We are looking for a creative and rigorous Research Scientist to focus on tailoring world models for effective use in policy learning and evaluation for autonomous vehicles. In this role, you will be at the heart of research efforts that bridge perception-driven environment models and the training of intelligent decision-making policies. Your work will ensure that learned world models can serve as faithful, controllable, and informative substrates for safe and robust policy optimization and evaluation. Responsibilities • Develop and refine world models that support realistic and diverse counterfactual reasoning, scenario generation, and policy rollout. • Ensure that world models are compatible with and useful for reinforcement learning, imitation learning, and offline policy evaluation techniques. • Design methods to synthesize high-risk or edge-case scenarios from world models, enabling robust stress-testing of autonomous policies. • Explore techniques such as latent-space simulation, world model distillation, differentiable simulation, and closed-loop evaluation to improve policy development and evaluation pipelines. • Partner with researchers in world modeling, planning, and safety evaluation to co-develop aligned architectures and learning objectives to ensure that learned models accurately capture agent-environment dynamics relevant to long-horizon planning and safety-critical decision-making. • Publish high-quality research and contribute to the community through open-source tools, benchmarks, and conference participation. Requirements • PhD in Computer Science, Robotics, Machine Learning, or a related field. • Strong background in at least two of the following areas: World models or model-based reasoning in dynamic environments, World model adaptation and fine-tuning, Offline RL or imitation learning, Model-based reinforcement learning (MBRL), Simulation-to-reality transfer, or Policy evaluation and safety assurance. • A track record of high-quality publications in ML or robotics venues (e.g., ICML, ICLR, NeurIPS, CoRL, RSS). • Familiarity with latent dynamics models (e.g., Dreamer, PlaNet, MuZero). • Understanding of uncertainty modeling, generalization, and robustness in learned environments. • Experience evaluating autonomous vehicle policies in simulation and real-world settings. • Experience in building or applying models for downstream evaluation of autonomous systems. • Proficiency in Python and ML frameworks (e.g., PyTorch, JAX). Benefits • 401(k) eligibility • various paid time off benefits, such as vacation, sick time, and parental leave • annual cash bonus structure Apply tot his job