# Don't Get Yourself into Trouble! Risk-aware Decision-Making for Autonomous Vehicles

**Authors:** Kasra Mokhtari, Alan R. Wagner  
**arXiv:** [2106.04625](https://arxiv.org/abs/2106.04625)  
**Date:** 2021-06-08  
**Summary page:** https://www.summarizepaper.com/en/paper/2106.04625v1/

## Key points

- Authors Kasra Mokhtari and Alan R. Wagner focus on risk prediction in autonomous vehicles.
- They highlight the importance of accurately forecasting risky situations for safe operation.
- Two key components of risk are identified: probability of an undesirable outcome and estimation of severity.
- Their approach uses a trained deep reinforcement learning model for navigating through crowds.
- The framework integrates high-level risk-based path planning with low-level control based on reinforcement learning principles.
- Rigorous evaluations using simulations like CARLA validate the methodology's effectiveness.
- The research aims to enhance safety measures by enabling proactive avoidance and response to risky scenarios in real-time.

## Summary

In their paper titled "Don't Get Yourself into Trouble! Risk-aware Decision-Making for Autonomous Vehicles," authors Kasra Mokhtari and Alan R. Wagner explore the crucial aspect of risk prediction in autonomous vehicles. They emphasize the importance of accurately forecasting risky situations to ensure safe operation. Building upon their previous research, Mokhtari and Wagner identify two key components that characterize risk: the probability of an undesirable outcome and an estimation of its severity. Their novel approach leverages a trained deep reinforcement learning model for navigating through crowds, serving as the foundation for a comprehensive risk-based decision-making framework tailored specifically for autonomous vehicles. This framework seamlessly integrates high-level risk-based path planning with low-level control based on reinforcement learning principles. Through rigorous evaluations using high-fidelity simulations like CARLA, the authors validate the effectiveness of their methodology and showcase promising outcomes. Their approach has the potential to significantly enhance safety measures by enabling proactive avoidance and response to risky scenarios in real-time. Overall, this innovative research contributes towards advancing autonomous vehicle capabilities and shaping future technology towards greater reliability and safety standards through prioritizing risk-aware decision-making strategies.