À propos de ce poste Research Scientist, Robot Learning - Continual Adaptation chez Tri
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 advancing the state of the art in AI, robotics, driving, and material sciences.
The Vision
We envision a future with robots that help people in our real environments through interactions that allow robots to understand, adapt, and grow with us. We believe that robots should and will work well alongside people: helping where wanted, and ultimately enabling people to spend more time on the activities they enjoy most. To achieve this, robots need to be able to operate reliably in messy, unstructured environments. We must also create robots that people can understand, collaborate with, and rely on.
The Team
Our goal is to revolutionize the field of robotics, enabling long-horizon dexterous behaviors to be efficiently taught, learned, and improved over time in diverse, real-world environments with people.
Our team has deep cross-functional expertise across hardware, simulation, perception, controls, machine learning, and human-robot interaction. We measure our success in terms of the development of fundamental capabilities, as well as research impact through open-source software and publications. Come join us in developing technologies that help people thrive!
The Opportunity
We are seeking a creative and technically strong researcher to advance post-training methods for generative robot policies. This role focuses on enabling pretrained robot policies to adapt rapidly from experience and human feedback, improve through autonomous interaction and collaboration, and retain useful capabilities over time.
You will develop algorithms and frameworks for post-training and continual learning in robot foundation models, with the goal of allowing robots to improve after deployment without retraining from scratch.
Responsibilities
Design and implement methods for adapting generative robot policies using techniques such as reinforcement learning, imitation learning, agentic tool use, and related methods.
Develop methods that enable generative robot policies to improve through ongoing interaction with people and the environment, using signals such as demonstrations, interventions, preference feedback, learned rewards, and other human-driven signals.
Develop rollout-based learning and exploration strategies that enable robots to efficiently and safely gather useful experience, leveraging multimodal data from demonstrations, teleoperation, interventions, and on-policy rollouts. In both simulation and on real robot hardware.
Collaborate across teams in perception, control, foundation models, and ML infrastructure to deploy scalable and reproducible research systems.
Publish research outcomes and contribute to the open robotics and embodied AI communities.
Qualifications
Ph.D. or M.S. in Robotics, Machine Learning, Computer Vision, or a related field, or equivalent applied research experience.
Strong background in reinforcement learning, imitation learning, or closely related robot-learning methods.
Strong proficiency with deep learning frameworks such as PyTorch or JAX and robotics simulation environments such as IsaacLab or MuJoCo.
Experience with policy adaptation, continual learning, preference learning, or related learning problems in embodied settings.
Strong coding and experimental skills with an emphasis on reproducibility and rigorous evaluation.
Prior robotics experience with real-world hardware and ML-based robot deployments.
Publications in top-tier conferences such as CoRL, RSS, NeurIPS, ICLR, ICML, ICRA, or CVPR.
Bonus Qualifications
Experience with human-in-the-loop robot learning (e.g., DAgger), preference learning, reward learning, or RLHF/RLAIF.
Prior work on generative robot policies or VLAs such as π0.5, GR00T, or related models.
Experience building or managing robot data collection or replay infrastructure.
Familiarity with real-world robot platforms such as YAMs, Kinova, humanoids, or mobile manipulators.
