Machine Learning Engineer Intern - Scenario Simulation

Bridge the Sim-to-Real Gap: Develop and implement a translation layer that converts idealized simulator Ground Truth (GT) into realistic, "noisy" Bird’s-Eye View (BEV) embeddings.

Optimize Simulation Throughput: Research and implement a "shortcut" pipeline that bypasses slow image/LiDAR rendering to generate BEV features directly from state data.

Enable Closed-Loop Training: Integrate the translated BEV embeddings into a training pipeline to make synthetic data directly usable for planning models.

Support Reinforcement Learning: Create the infrastructure necessary for planning models to undergo self-play RL fine-tuning within the bridged BEV feature space.