Scene Understanding in Deep Learning-Based End-to-End Controllers for Autonomous Vehicles

Yang, Shun Kun; Wang, Wenshuo; Liu, Chang; Deng, Weiwen

doi:10.1109/tsmc.2018.2868372

Cited by 81 publications

(35 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Before the online stage, the weights of the critic were tuned during the offline stage. According to Algorithm 1, the last W N is obtained using least-squares method, as in (19). Once W N is found, (17) can be used for calculating W N −1 .…”

Section: Traffic Network Benchmarkmentioning

confidence: 99%

“…Han and co-authors proposed a deep reinforcement learning model to decide the traffic signals' duration based on the collected data from different sensors and vehicular networks [14]. Reinforcement learning was also proposed for decision making of intelligent vehicles [17], [18], [19]. The curse of dimensionality is the main problem of microscopic frameworks: in fact, because the model describes dynamics at the vehicle level, the state easily becomes extremely large, making optimization prohibitive.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Switching-Based Adaptive Dynamic Programming Method to Optimal Traffic Signaling

Liu

Baldi

et al. 2020

IEEE Trans. Syst. Man Cybern, Syst.

View full text Add to dashboard Cite

The work presented in this paper concerns a switching-based control formulation for multi-intersection and multi-phase traffic light systems. A macroscopic traffic flow modelling approach is first presented, which is instrumental to the development of a model-based and switching-based optimization method for traffic signal operation, in the framework of Adaptive Dynamic Programming (ADP). The main advantage of the switching-based formulation is its capability to determine both 'when' to switch and 'which' mode to switch on without the need to use the cycle-based average flow approximation typical of state-of-the-art formulations. In addition, the framework can handle different cycle times across intersections without the need for synchronization constraints and, moreover, minimum dwell-time constraints can be directly enforced to comply with minimum green/red times in each phase. Simulation experiments on a multi-intersection and multi-phase traffic light systems are presented to show the effectiveness of the method.

show abstract

Section: Traffic Network Benchmarkmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

A Switching-Based Adaptive Dynamic Programming Method to Optimal Traffic Signaling

Liu

Baldi

et al. 2020

IEEE Trans. Syst. Man Cybern, Syst.

View full text Add to dashboard Cite

show abstract

“…. , s i , a i ) of the agent, in order to make the strategy produces a fixed trajectory, that is, the action output is unique under the same state, so a deterministic strategy is adopted [28]. At the same time, in order to avoid the inability to learn, due to the inability of certain strategies to access other states, the learning method of different strategies is adopted.…”

Section: Update Functionmentioning

confidence: 99%

“…ii) The decision-making process has little correlation with rules, and the algorithm lacks interpretability. iii) Most researchers still use images as the sole input for end-to-end active planning, which leads to inadequate representation of the driving state that has to be obtained by the DRL network [20], [28], [29]. iv) The training environment is mostly used in open-source game engines such as TORCS and Carla to verify the feasibility of the principle; this makes it difficult for the trained DRL control strategy to be directly applied from the virtual training scene to the real scene [30]- [32].…”

Section: Introductionmentioning

confidence: 99%

DDPG-Based Decision-Making Strategy of Adaptive Cruising for Heavy Vehicles Considering Stability

et al. 2020

View full text Add to dashboard Cite

The decision-making system of intelligent vehicles is the core component of an advanced driving system for both passenger vehicles and commercial vehicles. Finding ways to improve decisionmaking strategies to suit the complex and unfamiliar environments is a standing problem for traditional rule-based methods. This paper proposes a semi-rule-based decision-making strategy for heavy intelligent vehicles based on the Deep Deterministic Policy Gradient algorithm. Firstly, according to the car-following characteristics, the problems of high dimensions and a large amount of data in vehicle action space and state space are solved by dimension reduction and interval reduction to accelerate the training process. Subsequently, an accurate three-axle vehicle load model is established to calculate the load transfer rate value and carry out active control to increase the roll stability of heavy vehicles at high-speed corners. Furthermore, the Deep Deterministic Policy Gradient algorithm is developed based on the reward function and update function to achieve adaptive cruise control objectives for heavy vehicles on different curvature roads. Finally, the effectiveness and robustness of the algorithm are verified through simulation experiments. INDEX TERMS Adaptive cruise control, autonomous driving control system, decision-making algorithm, deep reinforcement learning, heavy vehicle, vehicle stability.

show abstract

“…Nowadays, with the rapid development of artificial intelligence (AI) techniques, the explainability and transparency of the decisions made by machine learning algorithms are becoming increasingly important [20], especially when these approaches started to be implemented for safety-critical applications, such as autonomous vehicles [43] and medical diagnose [13], etc. However, mainstream classification algorithms, for example, deep neural networks (DNNs) [27] and support vector machines (SVMs) [9] are typical type of "black box" models.…”

Section: Introductionmentioning

confidence: 99%

A hierarchical prototype-based approach for classification

Ding

2019

Information Sciences

View full text Add to dashboard Cite

In this paper, a novel hierarchical prototype-based approach for classification is proposed. This approach is able to perceive the data space and derive the multimodal distributions from streaming data at different levels of granularity in an online manner, based on which it further identifies meaningful prototypes to self-organize and self-evolve its hierarchical structure for classification. Thanks to the prototypebased nature, the system structure of the proposed classifier is highly transparent, and its learning process is of "one pass" type and computationally lean. Its decision-making process follows the "nearest prototype" principle and is fully explainable. The proposed approach is capable of presenting the learned knowledge from data in an easy-tointerpret prototype-based hierarchical form to users, and is an attractive tool for solving large-scale, complex real-world problems. Numerical examples based on various benchmark problems justify the validity and effectiveness of the proposed concept and general principles.

show abstract

Scene Understanding in Deep Learning-Based End-to-End Controllers for Autonomous Vehicles

Cited by 81 publications

References 16 publications

A Switching-Based Adaptive Dynamic Programming Method to Optimal Traffic Signaling

A Switching-Based Adaptive Dynamic Programming Method to Optimal Traffic Signaling

DDPG-Based Decision-Making Strategy of Adaptive Cruising for Heavy Vehicles Considering Stability

A hierarchical prototype-based approach for classification

Contact Info

Product

Resources

About