Search papers, labs, and topics across Lattice.
Tsinghua AICHDSchool of Traffic & Transportation EngineeringState Key Laboratory of Intelligent Green Vehicle and MobilityApr 20, 2026arXiv:2604.17841This paper introduces Evasive Acceleration (EA), a novel, two-dimensional approach to quantify collision risk in autonomous driving by evaluating the minimum acceleration needed to avoid a collision in any direction. EA is evaluated on five open datasets including over 600 real crashes, demonstrating that it provides earlier and more informative warnings compared to TTC-based methods. The results show that EA captures more outcome-relevant information, leading to significant information gain when integrated with existing risk assessment methods.
Time-to-collision metrics miss critical collision risk information, but a new 2D acceleration-based metric anticipates collisions far better.
Most autonomous driving safety benchmarks use time-to-collision (TTC) to assess risk and guide safe behaviour. However, TTC-based methods treat risk as a one-dimensional closing problem, despite the inherently two-dimensional nature of collision avoidance, and therefore cannot faithfully capture risk or its evolution over time. Here, we report evasive acceleration (EA), a hyperparameter-free and physically interpretable two-dimensional paradigm for risk quantification. By evaluating all possible directions of collision avoidance, EA defines risk as the minimum magnitude of a constant relative acceleration vector required to alter the relative motion and make the interaction collision-free. Using interaction data from five open datasets and more than 600 real crashes, we derive percentile-based warning thresholds and show that EA provides the earliest statistically significant warning across all thresholds. Moreover, EA provides the best discrimination of eventual collision outcomes and improves information retention by 54.2-241.4% over all compared baselines. Adding EA to existing methods yields 17.5-95.5 times more information gain than adding existing methods to EA, indicating that EA captures much of the outcome-relevant information in existing methods while contributing substantial additional nonredundant information. Overall, EA better captures the structure of collision risk and provides a foundation for next-generation autonomous driving systems.
