PSI Benchmark Dataset for Pedestrian Intention and Driver Decision Modeling
A new benchmark dataset named PSI has been developed by researchers to capture pedestrian crossing intentions from the viewpoint of drivers, complete with human-generated textual explanations. This dataset facilitates various tasks such as predicting pedestrian intentions, modeling driver decisions, generating reasoning, and forecasting trajectories. Its purpose is to enhance autonomous driving systems by merging predictive capabilities with human-aligned, interpretable reasoning. Additionally, PSI offers standardized evaluation protocols for assessing causal and interpretable outcomes. This research has been made available on arXiv, bearing the identifier 2112.02604.
Key facts
- PSI is a benchmark dataset for pedestrian intention and driver decision modeling.
- It captures dynamic evolution of pedestrian crossing intentions from the driver's perspective.
- Annotations include human textual explanations for intention estimation and driving decisions.
- Supports tasks: pedestrian intention prediction, driver decision modeling, reasoning generation, trajectory forecasting.
- Enables causal and interpretable evaluation for autonomous driving systems.
- Published on arXiv with identifier 2112.02604.
Entities
Institutions
- arXiv