Ji Hyun Yang

Development of a performance-based approach for a rear-end collision warning and avoidance system for automobiles

Many threat assessment algorithms are based on a collection of threshold equations that predict when a collision is to occur. The fact that there are numerous algorithms suggests a need to understand the underlying principles behind the equation design and threshold settings. In this paper, we present a methodology to develop appropriate alerting thresholds based on performance metrics. This also allows us to compare different alerting algorithms. The method is a performance-based approach in state-space. and can thus be utilized in conjunction with any chosen alerting algorithm or sensor system. Using carefully prescribed trajectory models (which may include uncertainties), the performance tradeoff with and without an alert can be predicted for different states along the course of an encounter situation. This information can then be used to set appropriate threshold values for the desired alerting logic. The development of the threshold criteria for a rear-end collision warning system is given as an example. Though the approach given is presented as a threshold design tool, the methodology is self-contained as a threat assessment logic. The possibility exists to compute the performance measures on-the-fly from which alerting decisions can be made directly.

A Real-Time Monte Carlo Implementation for Computing Probability of Conflict

*† ‡ § In this paper, we present a method for computing the Probability of Conflict, PC, using a fast Monte Carlo implementation. Often, Monte Carlo simulations are associated with offline analysis or verification, and are thought of as too slow for real-time usage. However, we describe an implementation that allows for fast computation and can be used in certain online applications. Examples are given for aircraft and automobile conflict detection. Both pseudo-code and Matlab code are provided to help explain and disseminate the approach.

Training Simulation for Helicopter Navigation by Characterizing Visual Scan Patterns

INTRODUCTION Helicopter overland navigation is a cognitively complex task that requires continuous monitoring of system and environment parameters and years of training. This study investigated potential improvements to training simulation by analyzing the influences of flight expertise on visual scan patterns. METHODS There were 12 military officers who varied in flight expertise as defined by total flight hours who participated in overland navigation tasks. Their gaze parameters were tracked via two eye tracking systems while subjects were looking at out-the-window (OTW) and topographic Map views in a fixed based helicopter simulator. RESULTS Flight performance measures were not predicted by the expertise level of pilots. However, gaze parameters and scan management skills were predicted by the expertise level. For every additional 1000 flight hours, on average, the model predicted the median dwell will decrease 28 ms and the number of view changes will increase 33 times. However, more experienced pilots scanned more OTW than novice pilots, which was contrary to our expectation. A visualization tool (FEST: Flight and Eye Scan visualization Tool) to replay navigation tasks and corresponding gaze data was developed. Qualitative analysis from FEST revealed visual scan patterns of expert pilots not only looking ahead on the map, but also revisiting areas on the map they just flew over to retain confidence in their orientation. DISCUSSION Based on the analysis provided above, this work demonstrates that neurophysiological markers, such as eye movements, can be used to indicate the aspects of a trainees cognitive state that are useful for cuing an instructional system.

Multiple model estimation for improving conflict detection algorithms

We present a framework for improving conflict detection algorithms using a hybrid control paradigm. This allows us to separate the problem into two parts: state/mode estimation and threat prediction. Since the dynamic equations for a conflict can change discretely depending on the situation, we propose the use of multiple model (MM) estimators to predict the situation and ultimately improve threat assessment. We provide an example using two different MM estimators for a rear-end collision warning system. The estimators can be used to determine the scenario mode as well as improve the state estimates

Bayesian modeling of pilot belief and visual misperception in helicopter overland navigation

This paper aims to provide a framework to model human belief and misperception in helicopter overland navigation. Helicopter overland navigation is known to be a challenging cognitive task, and understanding the cognitive processes associated with it is non-trivial. Twelve military personnel participated in the study and statistical analysis showed that their gaze parameters can be predicted by their level of expertise. Some pilots showed common visual misperception during the navigation task, which can be explained by the following errors: 1) confusion between inference and evidence, 2) incorrect mutually exclusive assumptions on the data, and 3) biased sampling. Simulation results on two cases observed in the experiments are given. Quantitative differences in dynamic perceptions between a Bayesian agent and misperceiving humans are presented with the suggested modeling framework.

Autonomy Balancing in a Manned-Unmanned Teaming (MUT) Swarm Attack

In this paper, we describe a framework for developing an interactive feedback model of manned-unmanned teaming (MUT) operational mode selections for a broad spectrum of unmanned vehicle (UV) autonomy levels. Though the highest autonomy levels are within reach as technology continues to advance, lower level autonomy or human manual control will still be needed depending on mission scenarios and dynamic situations. Understanding when and how we change the autonomy level of MUT is critical to ensure system safety and to maximize system performance. Thus, we propose to integrate feedback from various human state variables (i.e., physiological and behavioral signals such as heart rate, skin conductance level, and postures) for estimating human workload and interest level and key task performance measures (accuracy and speed for assigned missions, task interaction) into MUT systems so that the MUT adapts its mode automatically as needed. We developed RESCHU-SA (Research Environment for Supervisory Control of Heterogeneous Unmanned Vehicles Swarm Attacks), a modified version of the RESCHU simulator originally developed at MIT. We designed a human-in-the-loop experiment to collect baseline data for varying levels of autonomy using the RESCHU-SA along with a physiological sensor BioHarness. Different levels of autonomy include 1) high level autonomy using an auction algorithm or nearest-neighbor assignment algorithm, 2) low level autonomy using manual assignment, and 3) interactive autonomy which allows operators to change between high and low autonomy level. The purpose of the research is to investigate the level of autonomy that should be given to unmanned vehicles (UVs) to successfully complete a mission using a MUT in a swarm attack scenario.

Modeling Peripheral Vision for Moving Target Search and Detection

INTRODUCTION Most target search and detection models focus on foveal vision. In reality, peripheral vision plays a significant role, especially in detecting moving objects. METHODS There were 23 subjects who participated in experiments simulating target detection tasks in urban and rural environments while their gaze parameters were tracked. Button responses associated with foveal object and peripheral object (PO) detection and recognition were recorded. In an urban scenario, pedestrians appearing in the periphery holding guns were threats and pedestrians with empty hands were non-threats. In a rural scenario, non-U.S. unmanned aerial vehicles (UAVs) were considered threats and U.S. UAVs non-threats. RESULTS On average, subjects missed detecting 2.48 POs among 50 POs in the urban scenario and 5.39 POs in the rural scenario. Both saccade reaction time and button reaction time can be predicted by peripheral angle and entrance speed of POs. Fast moving objects were detected faster than slower objects and POs appearing at wider angles took longer to detect than those closer to the gaze center. A second-order mixed-effect model was applied to provide each subjects prediction model for peripheral target detection performance as a function of eccentricity angle and speed. About half the subjects used active search patterns while the other half used passive search patterns. DISCUSSION An interactive 3-D visualization tool was developed to provide a representation of macro-scale head and gaze movement in the search and target detection task. An experimentally validated stochastic model of peripheral vision in realistic target detection scenarios was developed.

Validity Analysis of Vehicle and Physiological Data for Detecting Driver Drowsiness, Distraction, and Workload

This study aims to validate vehicle and physiological readouts for the assessment of a drivers state. According to the Korea Transportation Safety Authoritys report (2012), 22.5% of 1000 drivers have experienced a crash or near-crash caused by drowsy or distracted driving. In this paper, 2 simulated driving-environment experiments were designed. One experiment was performed to analyze various characteristics of drowsy drivers. Four subjects participated in 2 experimental sessions with 2 different sleep durations the day prior to the experiment (above 7 hours of sleep versus below 4 hours of sleep). Subjects were expected to drive on a highway road at 80 km/h. Another experiment was designed to analyze the characteristics of distracted or high workload drivers. Sixteen subjects participated in 1 experimental session requiring different levels of distraction or difficulty (e.g., driving while conducting a secondary task versus driving only). Subjects were expected to drive on a downtown road at 40 or 60 km/h depending on the level of difficulty. The present study indicated that vehicle and physiological data have the potential to assess drowsiness, distraction, and high workload driving. The vehicle and physiological data validated in this study will be incorporated into a drivers state-assessment algorithm in the future.

Mode confusion in driver interfaces for adaptive cruise control systems

In this study, we developed a new driver interface for an adaptive cruise control system that suppresses mode confusion, based on a formal method for analyzing the correctness and succinctness of the interface and information. In the finite-state transition model for the interface, the states were grouped into a set of modes, each of which consists of the states that transition to the same state for the same user-triggered event. To test the mode awareness of our new interface and compare it with the mode awareness of a conventional interface, a set of human-in-the-loop experiments were conducted in a simulated environment using a driving simulator.

Takeover Requests in Simulated Partially Autonomous Vehicles Considering Human Factors

In the development of autonomous vehicles, the main focus of sensor research has been in relation to environmental perception, and only minimal work has focused on the human–vehicle interaction perspective. However, human factors need to be considered to ensure the safe operation of partially autonomous vehicles. This study briefly introduces a design methodology for the takeover request (TOR) time in National Highway Traffic Safety Administration Level 3 vehicles and compares four different TORs in a simulated environment based on human-in-the-loop experiments with various driving scenarios. A total of 30 drivers participated in the study, and the quantitative/qualitative data obtained show statistically significant differences between the four TOR thresholds. This study shows that the timing involved in the takeover can be obtained by using a performance-based approach considering human factors.