Tri13mo ago

Senior Machine Learning Researcher, Large Behavior Models & Diffusion Policy

Los AltosFull-timesenior

Data ScienceMachine LearningData & AI

0 views0 saves0 applied

Apply Now

Quick Summary

Overview

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.

Technical Tools

Data ScienceMachine LearningData & AI

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 Team

The Automated Driving Advance Development division at TRI focuses on enabling innovation and transformation at Toyota by building a bridge between TRI research and Toyota products, services, and needs. We achieve this through partnership, collaboration, and shared commitment. The Automated Driving Advance Development team is leading a new cross-organizational project between TRI and Woven by Toyota to research and develop a fully end-to-end learned automated driving / ADAS stack. This cross-org collaborative project is synergistic with TRI’s robotics divisions' efforts in Diffusion Policy and Large Behavior Models (LBM).

The Opportunity

We are looking for a Senior Machine Learning Researcher to join us in developing a state-of-the-art, pixels-to-action, end-to-end system for automated driving. As an expert in machine learning, you will contribute to designing and developing innovative models for our autonomy stack and deploying them on vehicle platforms to solve daily driving tasks and handle long-tail scenarios.

An ideal candidate has a strong track record of leading independent research efforts, preferably including mentoring and collaborating with less experienced students and researchers. You will help to drive our exploration into end-to-end learning approaches for automated driving, using large-scale sensor data directly for perception, planning, and prediction to overcome traditional "information bottlenecks." This includes expanding our successful Large Behavior Model (LBM) robotics efforts and Diffusion Policy (DP) research into the driving domain, designing scalable architectures, and integrating visual-language-action modalities. Beyond refining models for closed-loop driving on public roads and in simulation, you will also explore data quality filtering, transfer learning from diverse data sources, and edge deployment optimization. This work is part of Toyota’s global AI efforts to build a more coordinated global approach across Toyota entities.

Conduct ambitious research to advance the state-of-the-art in using new capabilities in generative AI (e.g., recent results in diffusion policy [1],[2]) for end-to-end perception, planning, and prediction in automated driving with a focus on computer vision as the primary sensing modality.

Research and implement scalable end-to-end architectures that process raw sensor data to generate vehicle trajectories, addressing the challenges of long-tail driving scenarios with low data coverage.

Prototype, validate, and iterate model architectures using imitation learning and large-scale data, ensuring robust performance across diverse scenarios.

Perform closed-loop evaluations in sensor simulations and real-world testing environments to rigorously assess model performance, stability, and scalability.

Explore multi-modal and language-conditioned models to broaden the applicability of end-to-end policies, using external data sources and transfer learning to enhance generalization.

Collaborate with researchers and engineers across TRI, Woven by Toyota, and Toyota’s global ecosystem to accelerate model deployment and evaluation in both controlled environments (closed-course) and public road driving.

Take the lead on writing and publishing research results in peer-reviewed venues.

A PhD or equivalent experience in a robotics-relevant or embodied-AI field such as Computer Science, Mathematics, Physics, or Engineering.

A consistent track record of publishing at high-impact conferences/journals (CVPR, ICLR, NeurIPS, ICML, CoRL, RSS, ICRA, ICCV, ECCV, PAMI, IJCV, etc.)

A consistent track record of independent research.

Demonstrated ability to independently formulate and complete a research agenda while collaborating across subject areas.

Experience training large-scale models, including foundation models (e.g., vision-language models, text-to-video models).

Proficiency in Python and C++ for implementing and evaluating research ideas.

Experience with robot motion planning techniques like trajectory optimization, sampling-based planning, and model predictive control, or experience with automated driving domains (e.g., perception, prediction, mapping, localization, planning, simulation).

Experience in developing production-level code for real-time operating systems.

Experience optimizing runtime-critical systems for Linux, UNIX-like real-time operating systems on automotive-grade compute platforms, and building safety-critical software architectures.