Research Scientist, Reinforcement Learning

DeepRoute; Fremont (CA); Presencial; USD 90.000 - 130.000; Jornada completa

Descripción de la vacante

DeepRoute in Fremont, California is seeking engineers to develop end-to-end autonomous driving systems using reinforcement learning. The ideal candidate will have experience training RL policies in safety-critical environments and expertise in modern RL algorithms like DQN and PPO. You will collaborate with teams to optimize driving behaviors, build parallel simulations, and improve safety. Knowledge of deep learning frameworks like PyTorch and experience with LLM fine-tuning are beneficial.

Formación

• Experience with massively parallel simulation environments.
• Knowledge of sim-to-real transfer techniques and domain randomization.
• Knowledge of model optimization (quantization, pruning) and CUDA is a plus.

Responsabilidades

• Train and deploy RL policies in closed-loop driving environments.
• Scale RL training using massively parallel simulation systems.
• Design and optimize reward functions for complex driving behaviors.
• Improve sim-to-real transfer for real-world robustness.
• Collaborate with cross-functional teams to integrate models into production systems.

Conocimientos

Proficiency in modern RL algorithms: DQN, PPO, SAC, TD3

Proficiency in modern RLHF algorithms: PPO, DPO, GRPO

Hands-on experience training reward models

Knowledge of distributed RL training at scale

Proficiency in Python, comfortable with C++

Proficiency in deep learning frameworks such as PyTorch

Herramientas

Ray

Horovod

Descripción del empleo

We are building next-generation end-to-end autonomous driving systems powered by reinforcement learning.

You will work on applying RL in closed-loop, safety-critical environments, leveraging large-scale simulation and real-world driving data to improve safety, comfort, and robustness.

• Train and deploy RL policies in closed-loop driving environments
• Scale RL training using massively parallel simulation systems
• Design and optimize reward functions for complex driving behaviors
• Improve sim-to-real transfer for real-world robustness
• Collaborate with cross-functional teams to integrate models into production systems

Core Technical Skills

• Proficiency in modern RL algorithms: DQN, PPO, SAC, TD3, etc.
• Proficiency in modern RLHF algorithms: PPO, DPO, GRPO, etc.
• Hands‑on experience training reward models and finetuning LLM/VLM/VLA
• Knowledge of distributed RL training at scale
• Proficiency with massively parallel simulation environmentsKnowledge of sim‑to‑real transfer techniques and domain randomization
• Proficiency in Python, comfortable with C++
• Proficiency in deep learning frameworks such as PyTorch
• Experience with distributed training frameworks (Ray, Horovod, etc.)
• Knowledge of model optimization (quantization, pruning) and CUDA is a plus
• Knowledge of traffic rules, driving behavior modeling

Preferred Qualifications

• Publications in top‑tier venues (ICML, NeurIPS, ICLR, CVPR, ICCV, ECCV, ICRA, IROS, etc.)
• Open‑source contributions to RL libraries or autonomous driving projects
• Previous experience with LLM fine‑tuning using RLHF
• Knowledge of safe RL, interpretable AI, or robustness techniques
• Familiarity with autonomous vehicle regulations and safety standards