Road Detection for Reinforcement Learning Based Autonomous Car

Holen, Martin; Saha, Rupsa; Goodwin, Morten; Omlin, Christian W.; Sandsmark, Knut Eivind

doi:10.1145/3388176.3388199

Cited by 9 publications

(8 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Training a self-driving car to stay on the road with Proximal Policy Optimization (PPO) has been demonstrated by Holen et al [8]. However, the transfer learning was from 2D simulation to 3D (and vise versa), while in this paper it is demonstrated for 3D simulation to a real vehicle.…”

Section: Related Workmentioning

confidence: 84%

Learning to Drive Fast on a DuckieTown Highway

Wiggers

Visser

2022

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

We train a small Nvidia AI JetRacer to follow the road on a small DuckieTown highway. In the real-world, roads do not always have the same appearance, so the system should not be trained on lane markings alone but on the complete view of the front camera. To make this possible, the system is trained in simulation using a recent reinforcement learning approach in an end-to-end fashion. This driving experience is then transferred to the circumstances encountered on a real track. Transfer learning is surprisingly successful, although this method is very sensitive to the details of the vehicle dynamics. We trained multiple models at different speeds and evaluated their performance both in simulation and in the real world. Increasing the velocity proves difficult, as the learned policy breaks down at higher speeds. The result is a small Nvidia AI JetRacer, which is able to drive around a DuckieTown highway, based on simulated experiences.

show abstract

Section: Related Workmentioning

confidence: 84%

Learning to Drive Fast on a DuckieTown Highway

Wiggers

Visser

2022

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

show abstract

“…Therefore, it has become a trend to introduce deep learning structures such as convolutional and recurrent neural networks in reinforcement learning. Deep Q-learning takes r + γmax a q(s , a ) as the target Q-value, and defines the loss function between the network output Q value and the target Q-value function shown in Equation (4).…”

Section: Deep Q-networkmentioning

confidence: 99%

“…With the continuous optimization of reinforcement learning theory and algorithms, the application of reinforcement learning algorithms for path planning has gradually become a research hotspot for solving path planning problems. Reinforcement learning does not require prior knowledge of complex environment models, which helps achieve a high level of human intelligence and becomes an attractive approach for path planning [2,3], unmanned driving [4], video games [5], robot control [6] and USV path planning [7]. Although traditional Q-learning algorithms [8] have better results for path planning, they still have slow convergence speed and cannot solve the real-world problems of large scale and high complexity [9].…”

Section: Introductionmentioning

confidence: 99%

An Improved Dueling Deep Double-Q Network Based on Prioritized Experience Replay for Path Planning of Unmanned Surface Vehicles

Zhu

et al. 2021

JMSE

View full text Add to dashboard Cite

Unmanned Surface Vehicle (USV) has a broad application prospect and autonomous path planning as its crucial technology has developed into a hot research direction in the field of USV research. This paper proposes an Improved Dueling Deep Double-Q Network Based on Prioritized Experience Replay (IPD3QN) to address the slow and unstable convergence of traditional Deep Q Network (DQN) algorithms in autonomous path planning of USV. Firstly, we use the deep double Q-Network to decouple the selection and calculation of the target Q value action to eliminate overestimation. The prioritized experience replay method is adopted to extract experience samples from the experience replay unit, increase the utilization rate of actual samples, and accelerate the training speed of the neural network. Then, the neural network is optimized by introducing a dueling network structure. Finally, the soft update method is used to improve the stability of the algorithm, and the dynamic ϵ-greedy method is used to find the optimal strategy. The experiments are first conducted in the Open AI Gym test platform to pre-validate the algorithm for two classical control problems: the Cart pole and Mountain Car problems. The impact of algorithm hyperparameters on the model performance is analyzed in detail. The algorithm is then validated in the Maze environment. The comparative analysis of simulation experiments shows that IPD3QN has a significant improvement in learning performance regarding convergence speed and convergence stability compared with DQN, D3QN, PD2QN, PDQN, PD3QN. Also, USV can plan the optimal path according to the actual navigation environment with the IPD3QN algorithm.

show abstract

“…The trained policy is transferred to realworld testing experiments for analyzing the perception and control perspective of the AC. Another group of researchers in [11] uses PPO for proposing a road detecting algorithm in an urban driving environment. They carry out experiments in a Udacity racing game simulator [12] as well as in a small OpenAI gym carracing-v0 environment [13].…”

Section: B Deep Reinforcement Learning Algorithms For Autonomous Drivingmentioning

confidence: 99%

Evaluating the Robustness of Deep Reinforcement Learning for Autonomous Policies in a Multi-agent Urban Driving Environment

Sharif¹,

Marijan²

2021

Preprint

View full text Add to dashboard Cite

Deep reinforcement learning is actively used for training autonomous driving agents in a vision-based urban simulated environment. Due to the large availability of various reinforcement learning algorithms, we are still unsure of which one works better while training autonomous cars in single-agent as well as multi-agent driving environments. A comparison of deep reinforcement learning in vision-based autonomous driving will open up the possibilities for training better autonomous car policies. Also, autonomous cars trained on deep reinforcement learning-based algorithms are known for being vulnerable to adversarial attacks, and we have less information on which algorithms would act as a good adversarial agent. In this work, we provide a systematic evaluation and comparative analysis of 6 deep reinforcement learning algorithms for autonomous and adversarial driving in four-way intersection scenario. Specifically, we first train autonomous cars using state-of-the-art deep reinforcement learning algorithms. Second, we test driving capabilities of the trained autonomous policies in single-agent as well as multi-agent scenarios. Lastly, we use the same deep reinforcement learning algorithms to train adversarial driving agents, in order to test the driving performance of autonomous cars and look for possible collision and offroad driving scenarios. We perform experiments by using vision-only high fidelity urban driving simulated environments.

show abstract

Road Detection for Reinforcement Learning Based Autonomous Car

Cited by 9 publications

References 6 publications

Learning to Drive Fast on a DuckieTown Highway

Learning to Drive Fast on a DuckieTown Highway

An Improved Dueling Deep Double-Q Network Based on Prioritized Experience Replay for Path Planning of Unmanned Surface Vehicles

Evaluating the Robustness of Deep Reinforcement Learning for Autonomous Policies in a Multi-agent Urban Driving Environment

Contact Info

Product

Resources

About