>

Algorithms for Decision Making

Mykel J. Kochenderfer, Tim A. Wheeler, and Kyle H. Wray

Textbook, MIT Press

A broad introduction to algorithms for decision making under uncertainty. We cover a wide variety of topics related to decision making, introducing the underlying mathematical problem formulations and the algorithms for solving them. Includes complete Julia implementations. Free to download!

Algorithms for Optimization

Mykel J. Kochenderfer and Tim A. Wheeler

Textbook, MIT Press

A comprehensive introduction to optimization with a focus on practical algorithms for the design of engineering systems. Includes complete Julia implementations.

Automotive Safety Validation in Simulation

Tim A. Wheeler

Ph.D. Thesis, Stanford University, Department of Aeronautics and Astronautics

Fully automated driving will require intelligent systems capable of understanding, reacting to, and interacting with the intricate complexities of the real world. With the onset of autonomous driving it becomes increasingly necessary to develop advanced tools for establishing trust in intelligent safety systems that act without or despite human input. This thesis presents novel contributions to simulation-based validation, including human driver behavior and sensor models, distributions over driving scenes, and a new technique for the accelerated validation of advanced automotive active safety systems such as autonomous vehicles.

Closed-Loop Policies for Operational Tests of Safety-Critical Systems

Jeremy Morton, Tim A. Wheeler, and Mykel J. Kochenderfer

IEEE Transactions on Intelligent Vehicles

Automotive manufacturers must establish confidence in the safety of their system through large-scale testing. Devoting resources to testing a system whose safety is uncertain is a risky proposition. We frame the test scheduling problem as a POMDP with belief-dependent rewards, show to to efficiently solve for optimal policies, and provide high-level takeaways for manufacturers and regulators.

Deep Stochastic Radar Models

Tim A. Wheeler, Martin Holder, Hermann Winner, and Mykel J. Kochenderfer

IEEE Intelligent Vehicles Symposium

We use deep learning to learn models over radar power return from very general and extendable inputs. The model has a raster grid and object list head, and leverages variational autoencoding, convolutional layers, and adversarial learning to produce a model which exhibits fundamental radar effects while remaining real-time capable.

Imitating Driver Behavior with Generative Adversarial Networks

Alex Kuefler, Jeremy Morton, Tim A. Wheeler, and Mykel J. Kochenderfer

IEEE Intelligent Vehicles Symposium

We simulate driver behavior using deep recurrent driver models using a very general lidar-based inputs. We extend Generative Adversarial Imitation Learning to the training of recurrent policies, and we demonstrate that our model outperforms 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.

Generalizable Intention Prediction of Human Drivers at Intersections

Derek Phillips, Tim A. Wheeler, and Mykel J. Kochenderfer

IEEE Intelligent Vehicles Symposium

New models are developed to predict whether a driver will turn left, right, or continue straight as they approach an intersection. We demonstrate prediction results up to 150m before the intersection. LSTMs are shown to outperform other conventional models, with mean cross-validated prediction accuracy averaging over 85% for both three- and four-way intersections.

Factor Graph Scene Distributions for Automotive Safety Analysis

Tim A. Wheeler, Philipp Robbel, and Mykel J. Kochenderfer

IEEE International Conference on Intelligent Transportation Systems

We use factor graphs to represent distributions over highway scenes which can be applied to arbitrary road geometries. The proposed method is shown to outperform the state of the art. We demonstrate the use of importance sampling to improve collision frequency estimates when scene generation models are used with behavior models in simulation.

Analysis of Microscopic Behavior Models for Probabilistic Modeling of Driver Behavior

Tim A. Wheeler, Philipp Robbel, and Mykel J. Kochenderfer

IEEE International Conference on Intelligent Transportation 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 compares several models from the literature for highway driving. We propose several validation metrics to measure the quality of resulting models and use these metrics to demonstrate that a class of Bayesian network models outperforms the state of the art.

POMDPs.jl: A Framework for Sequential Decision Making under Uncertainty

M. Egorov, Z. Sunberg, E. Balaban, T. Wheeler, J. Gupta, and M. Kochenderfer

Journal of Machine Learning Research

POMDPs.jl is an open-source framework for solving Markov decision processes (MDPs) and partially observable MDPs (POMDPs). POMDPs.jl allows users to express sequential decision making problems with minimal effort without sacrificing the expressive nature of POMDPs, making this framework viable for both educational and research purposes. It is written in the Julia language to allow flexible prototyping and large-scale computation that leverages the high-performance nature of the language. The associated JuliaPOMDP community also provides a number of state-of-the-art MDP and POMDP solvers and a rich library of support tools for help with implementing new solvers and evaluating the solution results.

Analysis of Recurrent Neural Networks for Probabilistic Modeling of Driver Behavior

Jeremy Morton, Tim A. Wheeler, and Mykel J. Kochenderfer

IEEE Transactions on Intelligent Transportation Systems

We study the effectiveness of recurrent neural networks in predicting acceleration distributions for car-following on highways. Long short-term memory recurrent networks are trained and used to propagate simulated vehicle trajectories over ten-second 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.

Learning Discrete Bayesian Networks from Continuous Data

Yi-Chun Chen, Tim A. Wheeler, and Mykel J. Kochenderfer

Journal of Artificial Intelligence Research

Real data often contains a mixture of discrete and continuous variables, but many Bayesian network structure learning and inference algorithms assume all random variables are discrete. We introduce a Bayesian discretization method for continuous variables with quadratic complexity and empirically show that the proposed method is superior to the state of the art. We also incorpate this into the structure learning process to discretize all continuous variables and simultaneously learn Bayesian network structures.

A Probabilistic Framework for Microscopic Traffic Propagation

Tim A. Wheeler, Philipp Robbel, and Mykel J. Kochenderfer

IEEE International Conference on Intelligent Transportation Systems

This paper describes a methodology for constructing human driving models 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.

Initial Scene Configurations for Highway Traffic Propagation

Tim A. Wheeler, Philipp Robbel, and Mykel J. Kochenderfer

IEEE International Conference on Intelligent Transportation Systems

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.