About the job
Avride builds autonomous vehicles and delivery robots across the United States, using shared technologies to advance both areas. The company’s approach combines self-driving cars and delivery robots under one roof, aiming to push the boundaries of autonomy.
The Research Engineer Intern position is based in Austin, TX. This internship centers on hands-on work with large-scale driving datasets, focusing on the intersection of research and engineering. Interns join the ML Prediction and Planning team, which develops machine learning models that help vehicles interpret their surroundings and make safe, effective decisions. The team’s work includes predicting the behavior of other road users and generating driving paths in complex, dynamic environments.
Interns collaborate closely with a senior researcher, tackling challenges that influence real-world driving performance. The role offers exposure to the full process of moving research ideas into working prototypes and testing them in safety-critical systems.
Internship Tracks and Focus
For Summer 2026, two internship tracks are available within the ML Prediction and Planning team. The responsibilities below describe the Autonomous Vehicles track.
- Applied Research Project: Lead a research effort to study how different model ensembling strategies affect the gap between open-loop (training) and closed-loop (simulation) results. Tasks include reviewing academic literature, developing hypotheses, and prototyping solutions using Python and machine learning frameworks such as PyTorch.
- Design Ensembling Strategies: Implement and test various ensembling methods, including blending models trained with different random seeds, combining checkpoints from separate training phases, and applying weighted or learned blending approaches.
- Run Controlled Experiments: Compare the outcomes of single-model and ensemble methods, examining the impact of seed and checkpoint diversity. Evaluate both open-loop metrics (like training and validation loss, accuracy) and closed-loop metrics (such as simulation performance, safety, and stability).
- Analyze Metric Alignment: Explore how improvements in open-loop metrics correspond to closed-loop results, and identify scenarios where ensembling enhances performance.
