Self-improvement for autonomous systems
- By Matt Leonard
- Aug 21, 2017
What if autonomous vehicles could teach themselves to become safer and more advanced?
Autonomous technologies have benefited from better sensing and computing, but they’re still not trustworthy enough for use in some situations. The Defense Advanced Research Projects Agency wants to bring the capabilities of machine learning to unmanned systems with its Assured Autonomy program.
Increased assurance means increased trust. Trust in autonomous system is one of the main considerations for Department of Defense in adopting the technologies more fully, according to a 2016 Defense Science Board report.
“Establishing trustworthiness at design time and providing adequate capabilities so that inevitable variations in operational trustworthiness can be assessed and dealt with at run time is essential, not only for operators and commanders, but also for designers, testers, policymakers, lawmakers, and the American public,” the report reads.
Currently, the assurances that autonomous systems operate safely and perform as expected are set when they are deployed. But advancements in machine learning mean that these cyber-physical systems can learn and their assurance can improve over time, something DARPA calls “continual assurance.”
To launch its Assured Autonomy program, DARPA recently held a proposers day, looking for design and analysis technologies that will provide continual assurance in learning-enabled autonomous systems. The focus of the program is on military applications, but the proposers day announcement said the findings could be relevant to other learning-enabled autonomous cyber-physical systems.
Paul Brubaker, the CEO and president of Alliance for Transportation Innovation, said the formation of this DARPA program was good news.
“This activity will go a long way to assuaging public concern regarding the safety of autonomous systems and self-driving vehicles,” Brubaker said in an email. “Kudos to DARPA for launching this project -- it’s the long pole in the tent to increasing the comfort to self-driving and turning the possibilities into reality.”
Matt Leonard is a reporter/producer at GCN.
Before joining GCN, Leonard worked as a local reporter for The Smithfield Times in southeastern Virginia. In his time there he wrote about town council meetings, local crime and what to do if a beaver dam floods your back yard. Over the last few years, he has spent time at The Commonwealth Times, The Denver Post and WTVR-CBS 6. He is a graduate of Virginia Commonwealth University, where he received the faculty award for print and online journalism.
Leonard can be contacted at email@example.com or follow him on Twitter @Matt_Lnrd.
Click here for previous articles by Leonard.