Paving the Way for Automated Driving SystemsJuly 2021
Topics: Systems Engineering (General), Modeling and Simulation, Operations Analysis, Software Testing
Automation already helps us with parking, staying in our lane, and emergency braking. But what if our vehicles could drive themselves—automatically detecting and responding to a changing environment? What if they could prevent the accidents caused by human error, offer the physically disabled greater mobility, and handle tasks that put humans at risk?
While the Society of Automotive Engineers has defined several levels of automated driving systems (ADS), we’re not yet at the top of the scale. Today, even the most automated vehicles deployed do not qualify as Level 5, or fully self-driving vehicles that can operate anywhere and in all conditions without human support.
Initiatives to realize this goal are underway, but challenges remain. MITRE is conducting research aimed at accelerating the deployment of all levels of ADS, in both civilian and military realms.
Much of that research takes place in our Mobile Autonomous Systems Experimentation (MASE) Lab. The lab features a Jeep Grand Cherokee outfitted with sensors, analytic and data recorders, and powerful computer processors.
“Teams use the Jeep and the lab’s other capabilities to evaluate existing technologies and prototype new ones,” says Zach LaCelle, a systems engineer who manages the MASE Lab. “They also identify solutions for emerging challenges and recommend the most promising ones to our sponsors and partners.”
Safety presents a significant barrier to the widespread adoption of autonomous systems. ADS will regularly need to respond to objects and situations they’ve never encountered before, which makes it tough to build in safety assurances using approaches designed for non-automated vehicles.
Improving Perception and Control
One line of research—being conducted for the U.S. Army—focuses on two capabilities vital for operations in off-road domains. One is perception: the vehicle’s ability to use sensors and cameras to detect and interpret what it encounters. The other, advanced control systems, involves the ability to identify appropriate navigation paths based on sensory inputs.
“On the perception side, we’re experimenting with novel approaches in machine learning to classify objects and inform the vehicle about where it’s safe to drive and where it isn’t,” LaCelle says.
“On the control side, one of the concerns is the vehicle often doesn’t drive very confidently—there’s a stop-and-start quality to its movements.” To address that, we’re considering how to use approaches developed for unmanned aircraft systems and apply them to ground vehicles.
“One of the great things about autonomy is that it’s cross-domain,” LaCelle says. “As approaches demonstrate effectiveness in the air, or in defense, we’re transferring the lessons learned and new technologies to surface transportation.”
Building Safety into the Entire Development Life Cycle
Automated systems hold great promise for improving safety by helping prevent crashes resulting from human error. Mechanical and system failures pose new safety risks, however. To mitigate them, Kent Hollinger, a Safety Management System (SMS) expert, has proposed an SMS framework specific to the ADS industry. He recommends incorporating safety management throughout the development life cycle—and beyond.
“Since ADS technology limits how and when drivers can interfere with the operation of the vehicle—especially in higher levels of autonomy—developers bear more responsibility for identifying safety issues and addressing them before harm results,” he says. Rather than relying on inspections and testing to find faults or safety concerns, Hollinger recommends a systematic risk management program that assures safety is designed into products.
“Proactive management of safety risk in the design, testing, demonstration, and deployment stages of ADS development can ensure continuous risk reduction,” he says.
Hollinger also recommends ongoing monitoring after the deployment of automated vehicles. “Continuous data analysis is critical for ADS technology since it’s likely unforeseen conditions will arise.”
Improving Data Capture for Safety Analysis
Today, some vehicles are already equipped with event data recorders (similar to aircraft black boxes) that capture rudimentary data from before and during a crash—things like acceleration and brake position.
“The good news is that, with ADS, we have a lot more data we can capture,” LaCelle says. “But we need an intelligent way to choose the most relevant data to log. It will be crucial to log the data based on the hazards the vehicle encounters, so we’re researching how to best do that. Our approach is system-agnostic and can be leveraged by all of industry.”
Still other research addresses the threat of cyber attacks on ADS perception systems; the unique problems associated with perception of pedestrians and other vulnerable road users; and how to build collision-avoidance algorithms into higher-level autonomous system behaviors.
“All of our research benefits from a huge back bench of talent,” LaCelle says. “In any effort we undertake, we can pull in experts in autonomy, systems engineering, SMS, and cybersecurity. That breadth of expertise is critical in the ADS space. It’s a main reason our sponsors seek us out.”
—by Marlis McCollum
And explore more at Focal Point: Surface Transportation