CISE Hosts Learning to Trust Autonomy Workshop to Foster Collaboration in Autonomous Systems
In March 2024, Roscoe, a Massachusetts State Police robot dog produced by Boston Dynamics, was shot three times while engaged in a standoff in Cape Cod. During the standoff, Roscoe cleared the building before police officers entered, demonstrating how robotics can help and protect humans.
The rapid advancement in autonomous systems and robotics has impacted industries from law enforcement to transportation to manufacturing. However, in real-world deployment, these systems face challenges due to public uncertainty and the potential unpredictable factors they will encounter. On November 22, 2024, the Boston University Center for Information and Systems Engineering (CISE) hosted the Learning to Trust Autonomy Workshop to foster this discussion among experts in autonomous systems. Organizers for the event included Professor (ECE) and CISE Director Ayşe Coşkun, Associate Professor and CISE faculty affiliate Wenchao Li (ECE, SE, CS), Associate Professor and CISE faculty affiliate Roberto Tron (ECE, ME), and Associate Professor Renato Mancuso (CS), who is also the advisor for the BU F1TENTH Autonomous Racing team.
“We created this event to ignite groundbreaking research and ideas in robotics and autonomous systems,” Coşkun stated during the opening remarks. “By bringing together faculty, students, and industry professionals, the event fosters collaboration and networking, empowering participants to solve complex and timely challenges in the field.”
With over 100 attendees from academia and industry, the workshop brought together diverse perspectives, creating a collaborative environment for participants to discuss autonomous systems. These conversations addressed developments in robotics, as well as the technological and social challenges of utilizing them outside of the testing environment.
“The goal of the workshop was to talk about how we can learn to trust systems that rely on data and machine learning in the real-world,” said Tron. “This question is not just academic because technology has a way of getting ahead of society. We’re bringing together different perspectives in autonomous systems as the only way to move forward and get society to trust these systems. Understanding autonomous systems better requires deep collaboration between academia, industry, and government. Only when there is alignment between the three can we expect society to fully trust these systems.”
The day culminated with an autonomous grand prix where student teams from four universities raced autonomous vehicles. “The F1Tenth Race is an exciting initiative and great challenge for young autonomy enthusiasts. It provides an opportunity for them to work across the whole stack of autonomous navigation from algorithm to hardware, and really test their skills at the limit in a real-world setting,” said Li. “We are thrilled to have brought this excitement to BU, and look forward to hosting more of these events in the future.”
Invited Talks
Ani Hsieh, University of Pennsylvania Associate Professor (Mechanical Engineering and Applied Mechanics), kicked off the workshop with her lecture titled “From Data-Driven to Physics Guided: A Pathway to More Efficient, Generalizable, and Robust Autonomy.” Hsieh’s talk addressed the environmental monitoring her group, the General Robotics, Automation Sensing, and Perception Lab (GRASP Lab), performs to collect data on ocean dynamics. Hsieh’s research looks at the impact of climate change through ocean dynamics and uses data and machine learning to model these patterns and better understand the ocean.
Mario Sznaier, Northeastern University Dennis Picard Trustee Professor (ECE), gave a lecture titled “Designing Provably Safe Controllers from Partial Information.” His talk examined the promise of learning-enabled autonomy, including higher efficiency and reducing human errors. Sznaier emphasized the importance of formalization in autonomous systems, which involves robotics safely operating around unknown dynamics. In real-world deployment, autonomous systems will encounter unpredictable factors and they must be able to continue operating and learn from these situations.
The invited talks concluded with a lecture from John Leonard, Massachusetts Institute of Technology Samuel C. Collins Professor of Mechanical and Ocean Engineering, titled “Spatial AI for Robots and Humans.” Leonard explained today’s new age of robotics learning from human imitation and his goal of using robotics to amplify human skills. Within transportation, robotics can aid new or elderly drivers by supplementing their capabilities. This technology has the potential to help support humans in the home, hospitals, factories, and beyond.
Industry Panel
The industry panel, moderated by Roberto Tron, featured Brendan Schulman, Vice President of Policy & Government Relations, Boston Dynamics; Ulrich Viereck, Senior Machine Learning Vision Engineer, Symbotic; and Mitchell Black, Lead Research Scientist, MIT Lincoln Lab. The panelists discussed the challenge of demonstrating to the public that autonomous systems can be trusted. A significant issue in integrating this technology into the real-world is that the public has a significantly lower tolerance for robotic mistakes than human error. As a result, failures cause big setbacks in gaining public trust. However, in driving for example, it’s statistically safer to drive in an autonomous vehicle than with a human driver. The panelists suggested different ways to gain public confidence in autonomous systems including showing the public more cases of robotic successes and public outreach to bolster trust.
F1TENTH Autonomous Grand Prix
The Learning to Trust Autonomy Workshop concluded with the F1TENTH Autonomous Grand Prix. Teams from Boston University, Lehigh University, University of Pennsylvania, and Massachusetts Institute of Technology built autonomous race cars and wrote their own software to compete in the event. Each team raced in fast-paced time trials and exciting head-to-head races.
The standings for the F1TENTH Autonomous Grand Prix were:
1st Place: Lehigh University, The Mountain Hawks
2nd Place: Boston University, F1TenthBU Acro
3rd Place: University of Pennsylvania, XLab
The Learning to Trust Autonomy Workshop is the second iteration of CISE’s Fall Workshop Series. The goal of these events is to catalyze collaboration, new research, and conversation around critical societal challenges within intelligent systems.
The inaugural CISE Fall Workshop in November 2023, AI for Understanding Earthquakes, drew experts from the United States and abroad to explore the intersection of artificial intelligence and earthquakes, leading to Associate Professor (ECE, CS, SE, CDS) and CISE faculty affiliate, Brian Kulis, being awarded an NSF grant to further research in this area. Additionally, this workshop led to Kulis, Professor and CISE Faculty Affiliate Prakash Ishwar (ECE, SE, CS), and Professor and CISE Faculty Affiliate Janusz Konrad (ECE) being awarded a Hariri Institute Focused Research Program (FRP) titled “AI for Understanding Earthquakes“.