Staff / Senior Machine Learning Engineer, Reinforcement Learning

Wayve•Sunnyvale, CA

20h•$311,850 - $389,400•Hybrid

About The Position

As a Senior / Staff Machine Learning Engineer in Wayve's AV Core organisation, you will advance reinforcement learning methods for end-to-end driving models. You will identify where learning from reward or feedback can improve beyond behavior cloning, then take promising ideas from design through large-scale experiments, rigorous evaluation, and integration into our best driving models. Driving Core team develops the learning methods that turn diverse driving data into robust closed-loop behavior. You will be a technical owner for reinforcement learning within the group, working closely with researchers and engineers across AV Core, Simulation, Evaluation, and Product Engineering. Success means producing measurable improvements in driving behavior. Core Model Safety team develops the core model competencies that enable safe, driverless operation. You will lead the technical direction and delivery of a learned emergency trajectory model for low-frequency, high-consequence maneuvers such as evasive steering and emergency braking. You will take the programme from problem definition through modelling, evaluation, integration, and evidence for deployment.

Requirements

A strong track record developing and experimentally validating reinforcement learning or closely related sequential decision-making methods on complex, high-dimensional problems.
Deep understanding of modern reinforcement learning fundamentals, including policy and value learning, off-policy learning, function approximation, distribution shift, and the failure modes of learned objectives.
Hands-on experience with behaviour cloning, reinforcement learning, or related methods.
Proficiency in Python and PyTorch, with strong software engineering practices and hands-on experience building reliable machine learning training and evaluation systems.
Excellent experimental judgement: able to turn an ambiguous behavioral problem into falsifiable hypotheses, useful metrics, disciplined ablations, and clear technical decisions.
Senior-level ownership and collaboration: able to lead a substantial technical area, work across research and engineering boundaries, and bring others along through clear written and verbal communication.

Nice To Haves

Experience with offline reinforcement learning, imitation learning, reward modeling, preference learning, or post-training of large neural policies.
Experience in autonomous vehicles, robotics, control, or another domain where policies interact with safety-critical physical systems, including an understanding of motion planning, vehicle dynamics, control, or collision avoidance.
Experience with closed-loop simulation, off-policy evaluation, uncertainty or calibration, and evaluation under rare or shifted conditions.
Experience training multimodal, transformer-based, or generative policy models at scale.
Proficiency in C++, CUDA, distributed training, or performance optimization for production machine learning systems.

Responsibilities

Shape and execute the reinforcement learning roadmap for Driving Core / Core Model Safety, selecting problems and methods against clear behavioral gaps and measurable success criteria.
Develop and evaluate post-behavior-cloning optimization methods, including offline and off-policy reinforcement learning as well as other reward-guided approaches; design the regularization, data strategy, and diagnostics needed to make policies reliably better.
Help improve the reward models and related learning signals used to train and evaluate driving policies, working with partner teams to strengthen their quality, scalability, and downstream usefulness.
Build robust training and experimentation workflows using large-scale driving data; diagnose distribution shift, objective misspecification, optimization instability, and data or evaluation bias.
Define evidence across offline metrics, open-loop tests, closed-loop simulation, and on-road evaluation, and distinguish genuine policy improvement from benchmark overfitting.
Productionize successful methods in the shared ML stack, communicate decisions and results clearly, and raise the technical bar through design reviews, code reviews, and mentoring.