Senior Algorithm Reinforcement Learning Engineer

该职位来源于猎聘 Role Mission Develop and productionize reinforcement learning methods for autonomous-driving decision-making, planning, simulation agents, or related AD applications. Convert research prototypes into measurable, reproducible, and safely deployable capabilities with clear benefits over existing baselines.

Key Responsibilities

• Identify AD problems where reinforcement learning offers a measurable advantage over supervised, imitation-learning, optimization-based, or rule-based approaches.
• Formulate state, action, reward, constraint, and termination definitions for driving-related learning problems.
• Develop and evaluate offline RL, online RL, model-based RL, imitation-learning-assisted RL, or hybrid planning approaches according to project needs.
• Build scalable training and evaluation workflows using simulation, historical driving data, and distributed compute resources.
• Analyze reward hacking, distribution shift, unstable training, poor generalization, and simulation-to-real discrepancies.
• Define multi-dimensional evaluation covering safety, comfort, progress, rule compliance, interaction quality, and rare-scenario robustness.
• Work with planning and active safety teams to integrate learned policies with runtime constraints, trajectory validation, and fallback mechanisms.
• Document experimental assumptions and results, maintain reproducible model artifacts, and support production validation and release decisions.

  Required

  Qualifications

• Bachelor’s degree or above in Computer Science, Machine Learning, Robotics, Control Engineering, or a related discipline.
• 5+ years of algorithm or machine learning experience, including substantial hands-on reinforcement learning development.
• Strong understanding of MDP/POMDP formulation, value-based and policy-based methods, offline or online RL, and evaluation under distribution shift.
• Strong Python skills and experience with PyTorch or a comparable deep learning framework.
• Experience conducting reproducible experiments and comparing learned policies against meaningful non-RL baselines.
• Experience using simulation or logged data for policy training and evaluation.
• Ability to translate research results into engineering requirements, measurable acceptance criteria, and production integration plans.

  Preferred

  Qualifications

• Advanced degree in Machine Learning, Robotics, Control, or a related field.
• Experience applying RL to autonomous driving, robotics, motion planning, traffic-agent simulation, or sequential decision-making.
• Experience with offline RL, constrained RL, safe RL, model-based RL, world models, or imitation-learning-assisted training.
• Experience integrating learned policies with C++ production systems or real-time planning stacks.
• Experience with closed-loop simulation and systematic safety-critical regression evaluation. Relevant publications are welcome, but production deployment and measurable engineering impact are equally valued.