A5037644321- Autonomous Vehicle Technology and Safety
- Traffic control and management
- Reinforcement Learning in Robotics
- Traffic Prediction and Management Techniques
- Adversarial Robustness in Machine Learning
- Traffic and Road Safety
- Anomaly Detection Techniques and Applications
- Speech Recognition and Synthesis
- Vehicular Ad Hoc Networks (VANETs)
- Vehicle emissions and performance
- Gaussian Processes and Bayesian Inference
- Music and Audio Processing
- Advanced Neural Network Applications
- Probability and Risk Models
- Indoor and Outdoor Localization Technologies
- Topic Modeling
- Robotic Path Planning Algorithms
- Time Series Analysis and Forecasting
- Human-Automation Interaction and Safety
- Simulation Techniques and Applications
- Robotics and Sensor-Based Localization
- Statistical Methods and Bayesian Inference
- Transportation Planning and Optimization
- Software Reliability and Analysis Research
- Speech and Audio Processing
Carnegie Mellon University
2018-2025
Ruijin Hospital
2024
Shanghai Jiao Tong University
2020-2024
Zhengzhou Children's Hospital
2022-2024
Zhengzhou University
2022-2024
Guangzhou University
2024
University of California, Los Angeles
2023-2024
Hebei Medical University
2023-2024
Jilin Medical University
2011-2023
Jilin University
2010-2023
End-to-end (E2E) models, which directly predict output character sequences given input speech, are good candidates for on-device speech recognition. E2E however, present numerous challenges: In order to be truly useful, such models must decode utterances in a streaming fashion, real time; they robust the long tail of use cases; able leverage user-specific context (e.g., contact lists); and above all, extremely accurate. this work, we describe our efforts at building an recog-nizer using...
Automated vehicles (AVs) must be thoroughly evaluated before their release and deployment. A widely used evaluation approach is the Naturalistic-Field Operational Test (N-FOT), which tests prototype directly on public roads. Due to low exposure safety-critical scenarios, N-FOTs are time consuming expensive conduct. In this paper, we propose an accelerated for AVs. The results can generate motions of other primary accelerate verification AVs in simulations controlled experiments. Frontal...
The safety of automated vehicles (AVs) must be assured before their release and deployment. current approach to evaluation relies primarily on 1) testing AVs public roads or 2) track with scenarios defined in a test matrix. These two methods have completely opposing drawbacks: the former, while offering realistic scenarios, takes too much time execute latter, though it can completed short amount time, has no clear correlation benefits real world. To avoid aforementioned problems, we propose...
Thus far, end-to-end (E2E) models have not been shown to outperform state-of-the-art conventional with respect both quality, i.e., word error rate (WER), and latency, the time hypothesis is finalized after user stops speaking. In this paper, we develop a first-pass Recurrent Neural Network Transducer (RNN-T) model second-pass Listen, Attend, Spell (LAS) rescorer that surpasses in quality latency. On side, incorporate large number of utterances across varied domains [1] increase acoustic...
Autonomous driving systems have witnessed significant development during the past years thanks to advance in machine learning-enabled sensing and decision-making algorithms. One critical challenge for their massive deployment real world is safety evaluation. Most existing are still trained evaluated on naturalistic scenarios collected from daily life or heuristically-generated adversarial ones. However, large population of cars, general, leads an extremely low collision rate, indicating that...
Abstract Defined traffic laws must be respected by all vehicles when driving on the road, including self-driving without human drivers. Nevertheless, ambiguity of human-oriented laws, particularly compliance thresholds, poses a significant challenge to implementation regulations vehicles, especially in detecting illegal behaviors. To address these challenges, here we present trigger-based hierarchical online monitor for self-assessment behavior, which aims improve rationality and real-time...
Driving style analysis plays a pivotal role in intelligent vehicle design. This paper presents novel framework for driving based on primitive patterns. To this end, Bayesian nonparametric approach hidden semi-Markov model (HSMM) is introduced to extract the patterns from muti-dimensional time-series data without prior knowledge of these For approach, hierarchical Dirichlet process (HDP) applied learn unknown smooth dynamical modes HSMM, called Two other types approaches (HDP-HMM and sticky...
In automatic speech recognition (ASR) what a user says depends on the particular context she is in. Typically, this represented as set of word n-grams. work, we present novel, all-neural, end-to-end (E2E) ASR system that utilizes such context. Our approach, which refer to Contextual Listen, Attend and Spell (CLAS) jointly-optimizes components along with embeddings During inference, CLAS can be presented phrases might contain-of-vocabulary (OOV) terms not seen during training. We compare our...
Misunderstanding of driver correction behaviors is the primary reason for false warnings lane-departure-prediction systems. We proposed a learning-based approach to predict unintended lane-departure and chances drivers bring vehicles back lane. First, personalized model lane-keeping behavior established by combining Gaussian mixture hidden Markov model. Second, based on this model, we developed an online model-based prediction algorithm forthcoming vehicle trajectory judge whether will act...
Accurately predicting and inferring a driver's decision to brake is critical for designing warning systems avoiding collisions. In this paper, we focus on intent in car-following scenarios from perception-decision-action perspective according his/her driving history. A learning-based inference method, using onboard data CAN-Bus, radar, cameras as explanatory variables, introduced infer drivers' braking decisions by combining Gaussian mixture model (GMM) with hidden Markov (HMM). The GMM used...
Existing neural network-based autonomous systems are shown to be vulnerable against adversarial attacks, therefore sophisticated evaluation of their robustness is great importance. However, evaluating the under worst-case scenarios based on known attacks not comprehensive, mention that some them even rarely occur in real world. Also, distribution safety-critical data usually multimodal, while most traditional and methods focus a single modality. To solve above challenges, we propose...
The past few years have witnessed an increasing interest in improving the perception performance of LiDARs on au-tonomous vehicles. While most existing works focus developing new deep learning algorithms or model ar-chitectures, we study problem from physical design perspective, i.e., how different placements multiple Li-DARs influence learning-based perception. To this end, introduce easy-to-compute information-theoretic sur-rogate metric to quantitatively and fast evaluate LiDAR placement...
The process to certify highly automated vehicles has not yet been defined by any country in the world. Currently, companies test on public roads, which is time-consuming and inefficient. We proposed accelerated evaluation concept, uses a modified statistics of surrounding importance sampling theory reduce time several orders magnitude, while ensuring results are statistically accurate. In this paper, we further improve concept using piecewise mixture distribution models, instead single...
Big data has shown its uniquely powerful ability to reveal, model, and understand driver behaviors. The amount of affects the experiment cost conclusions in analysis. Insufficient may lead inaccurate models, whereas excessive waste resources. For projects that millions dollars, it is critical determine right needed. However, how decide appropriate not been fully studied realm This paper systematically investigates this issue estimate much naturalistic driving (NDD) needed for understanding...
Long-tail and rare event problems become crucial when autonomous driving algorithms are applied in the real world. For purpose of evaluating systems challenging settings, we propose a generative framework to create safety-critical scenarios for specific task algorithms. We first represent traffic with series autoregressive building blocks generate diverse by sampling from joint distribution these blocks. then train model as an agent (or generator) search risky scenario parameters given...
Multi-agent navigation in dynamic environments is of great industrial value when deploying a large scale fleet robot to real-world applications. This paper proposes decentralized partially observable multi-agent path planning with evolutionary reinforcement learning (MAPPER) method learn an effective local policy mixed environments. Reinforcement learning-based methods usually suffer performance degradation on long-horizon tasks goal-conditioned sparse rewards, so we decompose the long-range...
Exiting from highways in crowded dynamic traffic is an important path planning task for autonomous vehicles (AVs). This can be challenging because of the uncertain motion surrounding and limited sensing/observing window. Conventional methods usually compute a mandatory lane change (MLC) command, but behavior (e.g., vehicle speed gap acceptance) should also adapt to conditions urgency exiting. In this paper, we propose reinforcement learning-enhanced highway-exit planner. The learning-based...
Dedicated Short Range Communication (DSRC) was designed to provide reliable wireless communication for intelligent transportation system applications. Sharing information among cars and between the infrastructure, pedestrians, or "the cloud" has great potential improve safety, mobility fuel economy. DSRC is being considered by US Department of Transportation be required ground vehicles. In past, their performance been assessed thoroughly in labs limited field testing, but not on a large...
At signalized intersections, vehicle speed profile plays a vital role in determining fuel consumption and emissions. With advances of connected automated technology, vehicles are able to receive predicted traffic information from the infrastructure real-time plan their trajectories fuel-efficient way. In this paper, an eco-driving model is developed for hybrid electric congested urban environment. The queuing process explicitly modeled by shockwave with consideration deceleration...