Machine Learning Engineer - End to End Autonomy

Staff Machine Learning Engineer – End to End Autonomy

Palo Alto, CA / Ann Arbor, MI / Product & Technology – AD/ADAS – Employee / hybrid

Woven by Toyota is enabling Toyota’s once-in-a-century transformation into a mobility company. Inspired by a legacy of innovating for the benefit of others, our mission is to challenge the current state of mobility through human‑centric innovation — expanding what “mobility” means and how it serves society.

Our work centers on four pillars: AD/ADAS, our autonomous driving and advanced driver assist technologies;
Arene, our software development platform for software‑defined vehicles;
Woven City, a test course for mobility; and Cloud & AI, the digital infrastructure powering our collaborative foundation. Business‑critical functions empower these teams to execute, and together, we’re working toward one bold goal: a world with zero accidents and enhanced well‑being for all.

At Woven by Toyota, we are at the forefront of developing advanced Machine Learning solutions for autonomous driving. Our team tackles groundbreaking challenges in designing state‑of‑the‑art neural networks, pioneering innovative end‑to‑end architectures, and advancing ML techniques in perception, prediction, and motion planning. We’re passionate about pushing the boundaries of autonomous systems through deep learning and optimization, particularly in complex 3D geometric computer vision scenarios.

We’re seeking passionate innovators and creative problem‑solvers eager to redefine mobility through cutting‑edge AI and robotics, contributing directly to shaping the future of self‑driving technology.

Woven by Toyota is developing a joint project between Toyota Research Institute (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 division’s efforts in Diffusion Policy and Large Behavior Models (LBM).

A technical lead who will work with the Lead engineers across Woven and TRI to help drive the strategy of the research development for end‑to‑end driving. The technical lead will also help bridge the connections between the research project and the production programs. This role requires strong communication skills and a collaborative mindset to navigate the joint nature of the Woven / TRI collaboration.

The applicant is expected to have a wide technical knowledge of the state‑of‑the‑art approaches in robotics/autonomous driving to define vision, scope, and to initiate longer‑term open‑research efforts.

• Drive the strategy and development of state‑of‑the‑art initiatives to accelerate end‑to‑end autonomous driving research, including onboard and offline applications, across Woven and TRI tech leads.
• Identify opportunities, resolve ambiguities, propose initiatives and execute to transfer successful research efforts to the production path.
• Lead the design, development, experimentation and benchmarks of ML models for e2e autonomous driving, ranging from data strategy, multistage training approaches, model selection and experimentation, as well as evaluation methods and deployment.
• Identify opportunities across the training stack and initiate efforts to increase the scalability of ML pipeline to support the training of large foundation models, and to optimize edge deployment needs of sota architectures.
• Collaborate, across time zones, with cross‑functional teams such as Data, Perception, Simulation, TRI Research teams to define interfaces and requirements for an end‑to‑end stack, and to influence technical decisions across the partnership with Woven and TRI to drive innovation.
• Initiate, lead and mentor on ML best practices across teams.

• 7+ years of professional experience with machine learning applications or applied science
• MS or higher degree in CS/CE/EE, or equivalent industry experience
• Hands‑on experience with recent breakthroughs in generative AI for robotics and/or autonomous driving, such as large e2e behavior models, foundation models, world models, multimodal transformer architectures, pre‑training and efficient fine‑tuning,…