Object Importance Estimation using Counterfactual Reasoning for Intelligent Driving

• DOI: 10.48550/arxiv.2312.02467
• Publication Date: 2023-01-01
• Authors: Pranay Gupta; Abhijat Biswas; Henny Admoni; David Held
• Keywords: Hoist (device)

The ability to identify important objects in a complex and dynamic driving environment is essential for autonomous agents make safe efficient decisions. It also helps assistive systems decide when alert drivers. We tackle object importance estimation data-driven fashion introduce HOIST - Human-annotated Object Importance Simulated Traffic. contains scenarios with human-annotated labels vehicles pedestrians. additionally propose novel approach that relies on counterfactual reasoning estimate an object's importance. generate by modifying the motion of ascribe based how modifications affect ego vehicle's driving. Our outperforms strong baselines task HOIST. perform ablation studies justify our design choices show significance different components proposed approach.