A simulation method to train autonomous vehicles to navigate complex urban scenes


The images above represent the trajectory of a vehicle making a sudden lane change. Our research on behavior-guided autonomous driving and simulation can predict these rare and interesting behaviors. Credit: Mavrogiannis, Chandra & Manocha.

In recent years, many companies, research organizations and academic institutions around the world have tried to develop safe and reliable autonomous vehicles. To be deployed on a large scale, however, these vehicles must be able to travel on a wide variety of roads and environments, without colliding with other vehicles, pedestrians, bicycles, animals or nearby obstacles.

Researchers at the University of Maryland recently developed a new technique that could improve the efficiency of simulators currently used to train models for autonomous vehicle navigation. This technique, presented in an article published in IEEE Letters on Robotics and Automationbuilds on their previous research focused on autonomous vehicle navigation.

“While there is currently a lot of interest in autonomous navigation for self-driving cars, current AI methods used for navigation do not take into account the behavior of human drivers or other self-driving vehicles on the road” , said Professor Dinesh Manocha, who supervised this research project, told TechXplore. “The objectives of our work are to develop robust technologies capable of detecting and classifying the behaviors of other traffic agents (e.g. vehicles, buses, trucks, bicycles, pedestrians) and to use these behaviors to guide the driving trajectories of autonomous vehicles.”

Generally, driving behaviors can be divided into two main categories, namely conservative or aggressive behaviors. As the name suggests, conservative drivers are more cautious and attentive, while aggressive drivers can be unstable and aggressive.

Accurately detecting these different driving modes can be very useful for autonomous vehicles, especially at critical moments (for example, when changing lanes or entering/exiting highways), as it allows them to to adapt their trajectories and their security measures accordingly. In the past, many teams have used simulation platforms to allow autonomous vehicles, as well as advanced driver assistance systems (ADAS), to accurately classify these driving behaviors.

“Autonomous driving navigation systems are typically trained in simulation before performing field tests,” Rohan Chandra, another researcher involved in the study, told TechXplore. “In our recent paper, we present a new behavior-based simulator that can emulate a wide variety of diverse behaviors seen in real-world traffic scenarios. This means that the underlying navigation system can be trained to handle traffic behavior. complex driving in real world traffic scenarios.”

The simulation technique introduced by the researchers is based on a model capable of classifying the driving behavior of other agents on the road. This model, called CMetric, analyzes the trajectories of other agents and then calculates them, using state-of-the-art computer vision tools.

A video explaining the broader ideas and impact of behavior-focused autonomous driving research conducted by the University of Maryland team. Credit: Mavrogiannis, Chandra & Manocha.

“Using CMetric, our behavior-guided simulator can generate agents with varied behaviors, resulting in mixed traffic scenarios,” Angelos Mavrogiannis, another researcher who conducted the study, told TechXplore. “The simulation of heterogeneous driving behaviors is a unique aspect of our work. We use a deep reinforcement learning policy based on DQN (Deep Q-Network), which we have integrated into our simulator.”

The driving behavior prediction model introduced by Mavrogiannis, Chandra, and Manocha can be integrated with a wide variety of state-of-the-art algorithms for vehicle navigation. This means that other teams around the world could use it to improve the training of their own models and improve overall performance.

Until now, most existing self-driving models have struggled to navigate complex urban environments. This includes roads with heavy traffic or with a high number of traffic lights, pedestrians and bicycles. The simulation technique developed by this team of researchers could ultimately help improve the performance of these models in these complex urban scenarios.

“Current autonomous driving systems are mainly applicable to highway traffic situations,” Chandra explained. “Our method, on the other hand, provides a new solution to simulate and evaluate autonomous driving technologies in complex or challenging urban scenes. This is even more important in terms of handling the harsh traffic conditions that are observed in cities. Asians, where the traffic density is higher, and many drivers do not follow the lanes of the traffic rules.Our simulator is the first step to generate these traffic patterns.

While primarily designed to be an algorithm learning tool, the simulation technique developed by the researchers can also be used to generate training datasets that also take into account driving behaviors and vehicle trajectories in complex urban environments. As part of their research, Mavrogiannis, Chandra and Manocha used these behavioral classification methods to create and analyze METEOR, a large-scale dataset containing dense, unstructured videos of intense traffic conditions. These videos were collected in India and then manually annotated by researchers to highlight rare or interesting driving behaviors, such as atypical road interactions and traffic violations.

In the future, the dataset published by the researchers could be used by other teams around the world to improve the navigation of autonomous vehicles and ADAS in congested and complex urban environments. The researchers now also plan to make the simulation technique they developed open source, so that other teams and companies can use it to train their own models and algorithms.

“We are currently developing better methods to classify traffic agent behavior using basic cameras (e.g., in smartphones) and use them to improve navigation of autonomous driving systems,” Chandra added. “These methods could also help a human driver in ADAS.”

A new, more realistic simulator will improve the safety of autonomous vehicles ahead of road tests

More information:
Angelos Mavrogiannis et al, B-GAP: behavior-rich simulation and navigation for autonomous driving, IEEE Letters on Robotics and Automation (2022). DOI: 10.1109/LRA.2022.3152594

Rohan Chandra et al, CMetric: a measure of driving behavior using centrality functions, 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (2021). DOI: 10.1109/IROS45743.2020.9341720

Rohan Chandra et al, Using Graph Theory Machine Learning to Predict Human Driver Behavior, IEEE Transactions on Intelligent Transportation Systems (2021). DOI: 10.1109/TITS.2021.3130218

Rohan Chandra et al, GraphRQI: Classifying Driver Behaviors Using Graph Spectra, 2020 IEEE International Conference on Robotics and Automation (ICRA) (2020). DOI: 10.1109/ICRA40945.2020.9196751

METEOR: Heterogeneous Conduct Dataset: gamma.umd.edu/meteor/

Provided by Science X Network

© 2022 Science X Network

Quote: B-GAP: A simulation method for training autonomous vehicles to navigate complex urban scenes (March 25, 2022) Retrieved March 25, 2022 from https://techxplore.com/news/2022-03-b-gap- simulation-method-autonomous-vehicles.html

This document is subject to copyright. Except for fair use for purposes of private study or research, no part may be reproduced without written permission. The content is provided for information only.


About Author

Comments are closed.