Robotics Intern, Post-Training and Continual Learning

Toyota Research Institute•Los Altos, CA

1d•Hybrid

About The Position

Toyota Research Institute’s Robotics division develops pretrained generalist policies for a broad range of tasks, environments, and robotic systems. The team studies how to adapt these broadly pretrained policies efficiently to target tasks while achieving the reliability, precision, and robustness required for real-world use. The intern will develop post-training methods that help general capabilities translate into dependable performance, including adaptation with limited additional data or interaction. Research interests include reinforcement learning, imitation learning, human-in-the-loop learning, simulation, world models, policy distillation, and large-scale robot learning. Research directions may include offline-to-online reinforcement learning; DAgger, imitation learning, and human-in-the-loop policy improvement; sim-to-real policy distillation and adaptation; policy improvement, planning, or data generation using world models; and data-efficient adaptation to new tasks and environments. Methods may be evaluated or deployed within internal research projects involving real-world industrial tasks. This is a paid, 12-week internship and a hybrid, in-office role. The expected California-based pay range at commencement is $45 to $65 per hour. TRI offers medical, dental, and vision insurance, as well as paid time off including holiday pay and sick time.

Requirements

Currently pursuing a Ph.D. in Computer Science, Machine Learning, Robotics, or a related field.
Research experience in robot learning, reinforcement learning, imitation learning, generative modeling, world models, or a related area.
Interest in open research problems involving large-scale machine learning grounded in physical systems.
Proficiency in Python and a deep-learning framework such as PyTorch.
Ability to collaborate effectively with researchers and engineers and communicate research findings clearly.
Submit a link to a Google Scholar profile and a complete publication list with the CV.

Nice To Haves

Experience with pretrained generalist policies, foundation models, or large-scale robot-learning systems.
Familiarity with offline or online reinforcement learning, DAgger, interactive learning, or human-in-the-loop methods.
Experience with simulation, sim-to-real transfer, policy distillation, or robotic manipulation.
Experience with learned world models, model-based reinforcement learning, or planning.
Publication record or interest in publishing at leading venues such as CoRL, NeurIPS, ICLR, ICML, RSS, ICRA, IROS, or related conferences and journals.
Interest in translating fundamental research into reliable methods that can be evaluated on real robotic systems and practical downstream tasks.

Responsibilities

Develop post-training methods that adapt pretrained generalist robot policies to new tasks, environments, and robotic systems.
Design and conduct rigorous experiments in simulation and, where appropriate, on real robotic systems.
Analyze results, communicate findings, and contribute to research publications and internal technical reports.