Learning End-to-end Autonomous Driving using Guided Auxiliary Supervision

Mehta, Ashish; Subramanian, Adithya; Subramanian, Anbumani

doi:10.1145/3293353.3293364

Cited by 26 publications

(18 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Besides, to solve the problem of the unstable learning process, the scholars devised a few network structures that can be used in complex environments. For example, Mehta et al [28] proposed multi-task learning from demonstration (MTLfD) framework that predicts visual affordances and action primitives and guides predictive driving commands through direct supervision. Sauer et al [29] presented a direct perception method that maps video inputs to intermediate representations and is adapted to autonomic guidance in sophisticated urban surroundings to reduce traffic accidents.…”

Section: Related Workmentioning

confidence: 99%

Deep Deterministic Policy Gradient Algorithm Based on Convolutional Block Attention for Autonomous Driving

et al. 2021

View full text Add to dashboard Cite

The research on autonomous driving based on deep reinforcement learning algorithms is a research hotspot. Traditional autonomous driving requires human involvement, and the autonomous driving algorithms based on supervised learning must be trained in advance using human experience. To deal with autonomous driving problems, this paper proposes an improved end-to-end deep deterministic policy gradient (DDPG) algorithm based on the convolutional block attention mechanism, and it is called multi-input attention prioritized deep deterministic policy gradient algorithm (MAPDDPG). Both the actor network and the critic network of the model have the same structure with symmetry. Meanwhile, the attention mechanism is introduced to help the vehicles focus on useful environmental information. The experiments are conducted in the open racing car simulator (TORCS)and the results of five experiment runs on the test tracks are averaged to obtain the final result. Compared with the state-of-the-art algorithm, the maximum reward increases from 62,207 to 116,347, and the average speed increases from 135 km/h to 193 km/h, while the number of success episodes to complete a circle increases from 96 to 147. Also, the variance of the distance from the vehicle to the center of the road is compared, and the result indicates that the variance of the DDPG is 0.6 m while that of the MAPDDPG is only 0.2 m. The above results indicate that the proposed MAPDDPG achieves excellent performance.

show abstract

Section: Related Workmentioning

confidence: 99%

Deep Deterministic Policy Gradient Algorithm Based on Convolutional Block Attention for Autonomous Driving

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Although there are the prototypes of autonomous vehicles currently tested on the regular streets, some of the challenges for the autonomous driving are not completely solved yet. Current challenges in autonomous vehicles development are sensor fusion [38,39,40,41], higher-level planning decisions [42,43,44,45,46], an end-to-end learning for autonomous driving [1,2,3,4,5,47,48,49], reinforcement learning for autonomous driving [5,50,51,52,53], and human machine interaction [54,55]. A systematic comparison of deep learning architectures used for autonomous vehicles is given in [56], a short overview of sensors and sensor fusion in autonomous vehicles is presented in [57].…”

Section: Related Workmentioning

confidence: 99%

“…The known solutions for end-to-end learning for autonomous driving [1,2,3,4,5] are developed mostly for the real vehicles, where the machine learning model used for inference is deployed on the high-performance computer, which is usually located in the trunk of the vehicle, or those solutions use very deep neural networks that are computationally expensive (e.g., using ResNet50 architecture in) [3]. However, our idea was to develop a significantly smaller solution, a light deep neural network, with similar performance during autonomous driving as known solutions, but using a smaller computational cost that will enable deployment on an embedded platform.…”

Section: Introductionmentioning

confidence: 99%

An End-to-End Deep Neural Network for Autonomous Driving Designed for Embedded Automotive Platforms

Kocic

Jovičić

Drndarevic

2019

Sensors

127

View full text Add to dashboard Cite

In this paper, one solution for an end-to-end deep neural network for autonomous driving is presented. The main objective of our work was to achieve autonomous driving with a light deep neural network suitable for deployment on embedded automotive platforms. There are several end-to-end deep neural networks used for autonomous driving, where the input to the machine learning algorithm are camera images and the output is the steering angle prediction, but those convolutional neural networks are significantly more complex than the network architecture we are proposing. The network architecture, computational complexity, and performance evaluation during autonomous driving using our network are compared with two other convolutional neural networks that we re-implemented with the aim to have an objective evaluation of the proposed network. The trained model of the proposed network is four times smaller than the PilotNet model and about 250 times smaller than AlexNet model. While complexity and size of the novel network are reduced in comparison to other models, which leads to lower latency and higher frame rate during inference, our network maintained the performance, achieving successful autonomous driving with similar efficiency compared to autonomous driving using two other models. Moreover, the proposed deep neural network downsized the needs for real-time inference hardware in terms of computational power, cost, and size.

show abstract

“…The known answers for start to finish learning for self-sufficient driving [4], [5], [14], [9], [30] are grown for the most part for the genuine vehicles, where the AI model utilized for deduction is conveyed on the superior PC, which is typically situated in the storage compartment of the vehicle, or those arrangements utilize profound neural systems that are computationally costly (e.g., utilizing ResNet50 design in) [14].…”

Section: Fig 2 Block Diagram Of Interface Between Ai Sensor and Actmentioning

confidence: 99%

“…To counter this problem, we propose to use the roof surface of the vehicle to generate power from solar energy and store it in Li-ion batteries and then use this power source to run SBC computers with the proposed CNN models. Based on the road types the appropriate CNN can be used instead of using one very complex, computationally expensive network [14].…”

Section: Fig 2 Block Diagram Of Interface Between Ai Sensor and Actmentioning

confidence: 99%

Low Memory Footprint CNN Models for end-to-end Driving of Autonomous Ground Vehicle and Custom Adaptation to Various Road Conditions.

2019

IJITEE

View full text Add to dashboard Cite

The Idea presented in this paper is to establish that different types of Convolution Neural Network (CNN) model are required to be used for different types of terrains, to optimally drive the vehicle. The Layers and the trainable variables required to be configured as per the road conditions. For simpler roads, lighter and simpler networks and for complicated road, a CNN model with heavier network is required. Existing CNN models were used to build the performance baseline with respect to simple and complicated road conditions. In the proposed CNN models, layers and trainable parameters are adjusted according to the road conditions. The objective of the proposed solution is to successfully drive with minimum number of convolution layers and trainable variables, fit for deployment on computer consuming less than 100 watts, without GPU, for moderate speed autonomous vehicle. These new sets of proposed CNN models are either equal or smaller in network size and trainable parameters, memory footprint and refresh rate compared to Alexnet and Nvidia model. The overall requirement of computational power, cost, and size is also reduced. In this paper, we recommend designing AV software with multiple CNN models to address different road conditions instead of on one Model for all road conditions. Finally, we establish that when the terrain is more complex and requires more features to be extracted then the CNN layers need to be tweaked to extract more features and the speed of the vehicle needs to be reduced.

show abstract

Learning End-to-end Autonomous Driving using Guided Auxiliary Supervision

Cited by 26 publications

References 15 publications

Deep Deterministic Policy Gradient Algorithm Based on Convolutional Block Attention for Autonomous Driving

Deep Deterministic Policy Gradient Algorithm Based on Convolutional Block Attention for Autonomous Driving

An End-to-End Deep Neural Network for Autonomous Driving Designed for Embedded Automotive Platforms

Low Memory Footprint CNN Models for end-to-end Driving of Autonomous Ground Vehicle and Custom Adaptation to Various Road Conditions.

Contact Info

Product

Resources

About