About the role#
This internship focuses on addressing the sim-to-real gap within autonomous driving simulation. You will work on translating idealized simulation data into realistic representations to improve planning models. This project aims to make synthetic data usable for training and enable self-play reinforcement learning for planning architectures.
What you'll do#
- Develop a translation layer to convert idealized simulator Ground Truth into realistic, noisy Bird’s-Eye View (BEV) embeddings.
- Research and implement a shortcut pipeline to generate BEV features directly from state data, bypassing slow image or LiDAR rendering.
- Integrate translated BEV embeddings into training pipelines to support planning models.
- Build the infrastructure required for planning models to undergo self-play reinforcement learning fine-tuning within the BEV feature space.
What you'll need#
- Strong foundation in deep learning, computer vision, and machine learning.
- Proficiency in Python and PyTorch.
- Experience with BEV or end-to-end autonomous driving architectures, including BEV generation, sensor fusion, and spatial transformation.
- Experience addressing the sim-to-real gap in autonomous systems or robotics.
- Nice to have: Experience with C++, end-to-end driving models, simulators (CARLA, IsaacSim, Gazebo), synthetic data generation, and ML infrastructure tools like Kubeflow or MLFlow.
- Currently pursuing a degree in Computer Science, Computer Engineering, Electrical Engineering, Mechanical Engineering, or Data Science.
Location & details#
- Location: Santa Clara, California
- Term: Summer 2026
- Modality: On-site
- Commitment: Full-time
- This is a paid internship.
