Tim Allan Wheeler

Analysis of Recurrent Neural Networks for Probabilistic Modeling of Driver Behavior

The validity of any traffic simulation model depends on its ability to generate representative driver acceleration profiles. This paper studies the effectiveness of recurrent neural networks in predicting the acceleration distributions for car following on highways. The long short-term memory recurrent networks are trained and used to propagate the simulated vehicle trajectories over 10-s horizons. On the basis of several performance metrics, the recurrent networks are shown to generally match or outperform baseline methods in replicating driver behavior, including smoothness and oscillatory characteristics present in real trajectories. This paper reveals that the strong performance is due to the ability of the recurrent network to identify recent trends in the ego-vehicles state, and recurrent networks are shown to perform as, well as feedforward networks with longer histories as inputs.

Imitating driver behavior with generative adversarial networks

The ability to accurately predict and simulate human driving behavior is critical for the development of intelligent transportation systems. Traditional modeling methods have employed simple parametric models and behavioral cloning. This paper adopts a method for overcoming the problem of cascading errors inherent in prior approaches, resulting in realistic behavior that is robust to trajectory perturbations. We extend Generative Adversarial Imitation Learning to the training of recurrent policies, and we demonstrate that our model rivals rule-based controllers and maximum likelihood models in realistic highway simulations. Our model both reproduces emergent behavior of human drivers, such as lane change rate, while maintaining realistic control over long time horizons.

Analysis of microscopic behavior models for probabilistic modeling of driver behavior

The accurate simulation and prediction of human behavior is critical in many transportation applications, including safety and energy management systems. Construction of human driving models by hand is time-consuming and error-prone, and small modeling inaccuracies can have a significant impact on the estimated performance of a candidate system. This paper presents a comparative evaluation of several probabilistic microscopic human behavior models from the literature trained on naturalistic data for free-flow, car following, and lane change context-classes on highways. We propose several metrics to quantify model quality and use these metrics to demonstrate that a new class of Bayesian network models outperforms the state of the art.

Generalizable intention prediction of human drivers at intersections

Effective navigation of urban environments is a primary challenge remaining in the development of autonomous vehicles. Intersections come in many shapes and forms, making it difficult to find features and models that generalize across intersection types. New and traditional features are used to train several intersection intention models on real-world intersection data, and a new class of recurrent neural networks, Long Short Term Memory networks (LSTMs), are shown to outperform the state of the art. The models predict whether a driver will turn left, turn right, or continue straight up to 150 m with consistent accuracy before reaching the intersection. The results show promise for further use of LSTMs, with the mean cross validated prediction accuracy averaging over 85% for both three and four-way intersections, obtaining 83% for the highest throughput intersection.

Learning discrete Bayesian networks from continuous data

Learning Bayesian networks from raw data can help provide insights into the relationships between variables. While real data often contains a mixture of discrete and continuous-valued variables, many Bayesian network structure learning algorithms assume all random variables are discrete. Thus, continuous variables are often discretized when learning a Bayesian network. However, the choice of discretization policy has significant impact on the accuracy, speed, and interpretability of the resulting models. This paper introduces a principled Bayesian discretization method for continuous variables in Bayesian networks with quadratic complexity instead of the cubic complexity of other standard techniques. Empirical demonstrations show that the proposed method is superior to the established minimum description length algorithm. In addition, this paper shows how to incorporate existing methods into the structure learning process to discretize all continuous variables and simultaneously learn Bayesian network structures.

A Probabilistic Framework for Microscopic Traffic Propagation

Probabilistic microscopic traffic models provide a statistical representation of interactive behavior between traffic participants. They are crucial for the validation of automotive safety systems that make decisions based on surrounding traffic. The construction of such models by hand is error-prone and difficult to extend to the complete diversity of human behavior. This paper describes a methodology for microscopic traffic model construction based on a Bayesian statistical framework connected to real-world data and applies it to learning models for free-flow, car following, and lane-change behaviors on highways. The evolution of traffic scenes is represented by a generative model learned for individual vehicles that captures their response to other traffic participants as well as the road structure. Our evaluation shows realistic behaviors over a four second horizon. A complete implementation is available online.

Factor graph scene distributions for automotive safety analysis

Automotive safety validation requires evaluation on a statistically representative set of roadway configurations and scene geometries. Scenes must be sampled from a statistical model representative of what actually occurs on roadways. This paper introduces a methodology for realistic scene model construction based on factor graphs that can be applied to arbitrary road geometries. Parameter learning for factor graphs is known to be convex. Experiments show that the proposed method is superior to the state of the art.

Initial Scene Configurations for Highway Traffic Propagation

Validation of automotive safety systems can be done by simulating millions of driving traces. It is important that the distribution of initial scenes for these driving traces be as representative of reality as possible so that safety risk can be estimated accurately. This paper presents a methodology for constructing probability distributions over initial highway scenes from which samples can be drawn for safety evaluation through simulation. A method for automated model construction based on a Bayesian statistical framework is introduced and applied to the NGSIM Highway 101 and Interstate 80 datasets. Four models of increasing complexity and fidelity are developed. A complete implementation is available online.

Deep stochastic radar models

Accurate simulation and validation of advanced driver assistance systems requires accurate sensor models. Modeling automotive radar is complicated by effects such as multipath reflections, interference, reflective surfaces, discrete cells, and attenuation. Detailed radar simulations based on physical principles exist but are computationally intractable for realistic automotive scenes. This paper describes a methodology for the construction of stochastic automotive radar models based on deep learning with adversarial loss connected to real-world data. The resulting model exhibits fundamental radar effects while remaining real-time capable.