Description
NVIDIA is seeking a hardworking intern with expertise in Reinforcement Learning and Multi-Modal World Simulation Model to propel the evolution of ML-centric autonomous driving and Physical AI solutions.
The focus of this role lies in model-centric RL, learning about world simulation models, and translating state-of-the-art (SOTA) algorithms into real-world applications, allowing vehicles to interpret, anticipate, and respond astutely in challenging dynamic contexts.
Responsibilities:
- Develop and refine multi-modal world models and integrate them into our simulation system.
- Train and evaluate self-supervised latent dynamics and sensor generation models for the joint tasks of trajectory prediction, goal-conditioned ego control, and sensor data synthesis. Explore and prototype hybrid architectures combining world models, generative (e.g., diffusion, flow matching) models, and policy gradients for realistic and robust simulation.
- Collaborate with End-to-End Driving Model teams to deploy world-model-based policies to simulated RL environments and accelerate the training of the driving systems.
- Contribute to system development for continuous learning and simulation adaptation (Sim2Real transfer).
Requirements:
- Pursuing PhD in Computer Science, Machine Learning, or a related field, with neural rendering, robotics, or simulation background.
- Strong understanding of reinforcement learning (policy gradients, actor-critic, offline RL).
- Familiarity with visual representation learning and 4D scene representation (NeRF, Gaussian Splatting, occupancy networks and contrastive, masked modeling, or generative world simulation) for world simulation.
- Experience building large-scale training pipelines with temporal consistency and simulation data replay.
- Publications or open-source contributions in RL, model-based control, or autonomous systems.
- Passion for developing learning systems that can “imagine” and plan in the real world.