Analyzing data for transportation systems using TACC's Rustler, XSEDE ECSS support
Published on February 13, 2017 by Faith Singer-Villalobos
In the next 10 years you are going to see some form of autonomous or connected vehicles on the streets. Natalia Ruiz-Juri, a research associate with The University of Texas at Austin's Center for Transportation Research (CTR) is fairly certain of this. She is one of many researchers at CTR and The University of Texas at Austin (UT Austin) who are studying the wide range of technical, social and policy aspects of connected and autonomous vehicle (CAV) technologies.
Fully autonomous vehicles or driverless cars are capable of sensing their environment and navigating without human input. They can detect surroundings using a variety of techniques such as radar, lidar, GPS, odometry, and computer vision. Similarly, connected vehicles (CVs) are vehicles that can exchange messages containing location and other safety-related information with other vehicles, and with devices affixed to roadside infrastructure.
CVs share information in the form of Basic Safety Messages (BSMs) with other vehicles and the infrastructure; these include vehicle position, speed and breaking status. Such real-time feedback and information exchange between vehicles is expected to greatly enhance safety, and it opens the door to several possibilities in traffic management.
For example, vehicles could talk to other vehicles that are much further ahead and get warned about congestion or dangerous conditions, thereby allowing a driver to make strategic decisions and take a different path.
Additionally, vehicles could also talk to infrastructure, such as an intersection light, which might be capable of tracking the number of vehicles passing through and potentially adjusting the signal timing plan accordingly. The advent of CVs would therefore have huge promise in improving traffic management and the overall utilization of transportation infrastructure, particularly if vehicle connectivity is considered along with automation.
While the basic goal of CVs, in particular, is safety — experts hypothesize up to 80 percent less accidents in the future — the data generated by CVs has an enormous potential to support transportation planning and operations.
THE BIG DATA PROBLEM
At this point researchers are still exploring diverse datasets. A number of connected vehicle test beds and autonomous vehicles test sites have been planned, or are already in place. Texas is part of one of the 10 US-Department of Transportation-designated autonomous vehicle proving grounds, and research sponsored by other agencies, such as TxDOT and the North Central Texas Council of Governments is also happening at UT Austin.
"The volume and complexity of CV data are tremendous and present a big data challenge for the transportation research community," Ruiz-Juri said. While there is uncertainty in the characteristics of the data that will eventually be available, the ability to efficiently explore existing datasets is paramount.
Ruiz-Juri and her colleagues, including Chandra Bhat, James Kuhr and Jackson Archer, were interested in exploring the most comprehensive data set released to date — the Safety Pilot Model Deployment (SPMD) data, produced by a study conducted by The University of Michigan Transportation Research Institute and the National Highway Traffic Safety Administration.
However, to get started they needed help using computational resources. They turned to the Texas Advanced Computing Center (TACC), also at UT Austin, and a key partner in the Extreme Science and Engineering Discovery Environment (XSEDE), the most advanced, powerful, and robust collection of integrated advanced digital resources and services in the world. Through XSEDE, Ruiz-Juri and team took advantage of the Extended Collaborative Support Services (ECSS) program, and the TACC experts within the program, to make these resources easier to use and to help more people use them.
TACC ECSS experts Weijia Xu and Amit Gupta were able to help Ruiz- Juri and her colleagues figure out how to use very large datasets on supercomputers like Rustler, TACC's experimental system for exploring new storage and data compute techniques and technologies.
Ruiz-Juri and her colleagues compared efforts to build scalable solutions for CV data analysis using Hive, an open-source data warehouse application that supports distributed queries. The data included approximately 2,700 cars, trucks and transit buses whose activities were logged through on-board sensors over a two month period.
"Hive is an ideal choice in this particular use case since it not only offers scalability and performance but also has a SQL-like interface," Xu said, referring to Structured Query Language used to manage data. "It is similar to PostgreSQL which the research team is already familiar with."
According to Ruiz-Juri, using Rustler is a huge time-saver because it lets them play with the data and see what it looks like without spending hours waiting for a query to complete.
As a researcher, Ruiz-Juri said one of the challenges she faces is not knowing which system to use on a particular model for a particular dataset. This is one of the many ways that Xu and Gupta were able to help. They developed an automated methodology to understand how each system is expected to perform based on the characteristics of the network. For this work they used Rustler, but soon they plan to move the data to Wrangler, an XSEDE-allocated resource.
"Natalia and her colleagues were trying to make sense of the data," Gupta said. "It's unfiltered data from real people capturing their movement patterns across the city. All of this data was sampled at 10 times per second — speed data, when a person used their brakes, when they used their windshield wipers etc — so it's a lot of information and nobody has completely figured out what to do with it. Natalia and her team are trying to validate, and in some cases possibly break through, some of the assumptions that they traditionally made in their field with respect to traffic patterns."
In addition to determining which system to use for which model, Xu and Gupta also helped Ruiz-Juri and colleagues create a friendly user interface to remove some of the hurdles of using a command line. If you don't have an interface, the researcher has to come up with something manually and they may not have time or funding to do that, especially when in exploration mode. "The interface gave us the opportunity to look at this data now instead of, say, two years down the line in the project," Ruiz-Juri said.
"The XSEDE ECSS program has been great for us," Ruiz-Juri said. "We get together and we talk about projects and research in general. Amit and Weijia have started to understand more about what we are doing, so for me the best part is not when we know what we want and they help us, but when they understand enough of what we're doing and can come up with new ideas on their own. We've been working together for over three years now on different projects."
The goal is to enable their research exploration by leveraging HPC tools and infrastructure, according to Gupta. Due to the scale of such resources available at TACC, they are able to iterate through their analysis cycle much quicker and converge towards conclusions faster. It also enables them to attempt new simulation experiments that would overload their computational resources or take prohibitively long to run.
"I enjoy working on this project very much," Gupta said. "It's one of my favorite projects. It's a very challenging and interesting application of computer science to a real world problem."
THE CHALLENGE OF TRANSFORMATIVE TECHNOLOGIES
One of the challenges with research in this field is that connected and autonomous vehicles can be disruptive, according to Ruiz-Juri. How do we anticipate what's going to happen in the future when this type of technology can change not only transportation system performance, but also travel choices and behavior?
How are people going to react to this technology? Are they going to purchase more cars, fewer cars? Are they going to travel further? Are they not going to care about travel time any longer so they move further away from downtown?
Researchers want to understand how they need to modify existing models so that they can consider all these complex, interrelated impacts, when assessing the effects of CAV technologies into the future. Advanced models require significant computational resources, and TACC experts have already supported CTR in the use of HPC for their simulations.
"It would have been very hard without the help of Amit or Weijia to be able to have visibility and access to HPC from the interface of a preexisting code that we use for modeling, and which may be central to future research in CVs. They helped us a lot in terms of how to access the systems, how to set up log-in, writing scripts, authentication, creating accounts, and much more," Ruiz-Juri said.
Using models to test all of the hypotheses and questions can transform the way we think about living and travelling.
"I think that vehicle connectivity is going to happen relatively soon and it's going to make travel safer — it's something to look forward to," Ruiz-Juri said. "It also gives us the opportunity to collect a lot of data so we can look at operating transportation systems differently. It has huge potential for safety and traffic operations. Automated vehicles are an exciting possibility that can truly transform how we travel, and lead to major changes in lifestyle choices and decisions."
Natalia Ruiz Juri of the Center for Transportation Research, The University of Texas at Austin
Preliminary visualization of trip-level data after processing on Rustler.
A web-based interface to run large-scale advanced transportation models in TACC's HPC resources.