Many serious automobile accidents could be avoided if drivers were warned of impending crashes before they occur. Creating such warning systems by hand, however, is a difficult and time-consuming task. This paper describes three advances toward evolving neural networks with NEAT (NeuroEvolution of Augmenting Topologies) to warn about such crashes in real-world environments. First, NEAT was evaluated in a complex, dynamic simulation with other cars, where it outperformed three hand-coded strawman warning policies and generated warning levels comparable with those of an open-road warning system. Second, warning networks were trained using raw pixel data from a simulated camera. Surprisingly, NEAT was able to generate warning networks that performed similarly to those trained with higher-level input and still outperformed the baseline hand-coded warning policies. Third, the NEAT approach was evaluated in the real world using a robotic vehicle testbed. Despite noisy and ambiguous sensor data, NEAT successfully evolved warning networks using both laser rangefinders and visual sensors. The results in this paper set the stage for developing warning networks for real-world traffic, which may someday save lives in real vehicles.

Many serious automobile accidents could be avoided if drivers were warned of impending crashes before they occurred. In this paper, a vehicle warning system is evolved to predict such crashes in the RARS driving simulator. The NeuroEvolution of Augmenting Topologies (NEAT) method is first used to evolve a neural network driver that can autonomously navigate a track without crashing. The network is subsequently impaired, resulting in a driver that occasionally makes mistakes and crashes. Using this impaired driver, a crash predictor is evolved that can predict how far in the future a crash is going to occur, information that can be used to generate an appropriate warning level. The main result is that NEAT can successfully evolve a warning system that takes into account the recent history of inputs and outputs, and therefore makes few errors. Experiments were also run to compare training offline from previously collected data with training online in the simulator. While both methods result in successful warning systems, offline training is both faster and more accurate. Thus, the results in this paper set the stage for developing crash predictors that are both accurate and able to adapt online, which may someday save lives in real vehicles.

Objective: The objective of this research study is to estimate the benefit to pedestrians if all U.S. cars, light trucks, and vans were equipped with an automated braking system that had pedestrian detection capabilities. Methods: A theoretical automatic emergency braking (AEB) model was applied to real-world vehicle-pedestrian collisions from the Pedestrian Crash Data Study (PCDS). A series of potential AEB systems were modeled across the spectrum of expected system designs. Both road surface conditions and pedestrian visibility were accounted for in the model. The impact speeds of a vehicle without AEB were compared to the estimated impact speeds of vehicles with a modeled pedestrian detecting AEB system. These impacts speeds were used in conjunction with an injury and fatality model to determine risk of Maximum Abbreviated Injury Scale of 3 or higher (MAIS 3þ) injury and fatality. Results: AEB systems with pedestrian detection capability, across the spectrum of expected design parameters, reduced fatality risk when compared to human drivers. The most beneficial system (time-to-collision [TTC] ¼ 1.5 s, latency ¼ 0 s) decreased fatality risk in the target population between 84 and 87% and injury risk (MAIS score 3þ) between 83 and 87%. Conclusions: Though not all crashes could be avoided, AEB significantly mitigated risk to pedestrians. The longer the TTC of braking and the shorter the latency value, the higher benefits showed by the AEB system. All AEB models used in this study were estimated to reduce fatalities and injuries and were more effective when combined with driver braking.

Modeling vehicle-pedestrian interactions in the road environment is essential to develop pedestrian detection and pedestrian crash avoidance systems. In this paper, one novel approach is proposed to estimate the vehicle-pedestrian encountering risk in the road environment based on a large scale naturalistic driving data collection. Considering the difficulty to record actual pedestrian crashes in the naturalistic data collection, the encountering risk is estimated by the chances for driver to meet with pedestrian in the roadway as well as the chances for the driver and pedestrian to get into a potential conflict. Effects of different scenarios consisting of road conditions, pedestrian behaviors, and pedestrian numbers on the risk levels are also evaluated, and significant results are provided.

Objective: Accounting for one fifth of all crashes and one sixth of all fatal crashes in the United States, intersection crashes are among the most frequent and fatal crash modes. Intersection advanced driver assistance systems (I-ADAS) are emerging vehicle-based active safety systems that aim to help drivers safely navigate intersections. The objective of this study was to estimate the number of crashes and number of vehicles with a seriously injured driver (Maximum Abbreviated Injury Scale [MAIS] 3+) that could be prevented or reduced if, for every straight crossing path (SCP) intersection crash, one of the vehicles had been equipped with an I-ADAS. Methods: This study retrospectively simulated 448 U.S. SCP crashes as if one of the vehicles had been equipped with I-ADAS. Crashes were reconstructed to determine the path and speeds traveled by the vehicles. Cases were then simulated with I-ADAS. A total of 30 variations of I-ADAS were considered in this study. These variations consisted of 5 separate activation timing thresholds, 3 separate computational latency times, and 2 different I-ADAS response modalities (i.e., a warning or autonomous braking). The likelihood of a serious driver injury was computed for every vehicle in every crash using impact delta-V. The results were then compiled across all crashes in order to estimate system effectiveness. Results: The model predicted that an I-ADAS that delivers an alert to the driver has the potential to prevent 0-23% of SCP crashes and 0-25% of vehicles with a seriously injured driver. Conversely, an I-ADAS that autonomously brakes was found to have the potential to prevent 25-59% of crashes and 38-79% of vehicles with a seriously injured driver. I-ADAS effectiveness is a strong function of design. Increasing computational latency time from 0 to 0.5 s was found to reduce crash and injury prevention estimates by approximately one third. For an I-ADAS that delivers an alert, crash/injury prevention effectiveness was found to be very sensitive to changes in activation timing (warning delivered 1.0 to 3.0 s prior to impact). If autonomous braking was used, system effectiveness was found to largely plateau for activation timings greater than 1.5 s prior to impact. In general, the results of this study suggest that I-ADAS will be 2-3 times more effective if an autonomous braking system is utilized over a warning-based system. Conclusions: This study highlights the potential effectiveness of I-ADAS in the U.S. vehicle fleet, while also indicating the sensitivity of system effectiveness to design specifications. The results of this study should be considered by designers of I-ADAS and evaluators of this technology considering a future I-ADAS safety test.