Researchers from the University of Illinois at Urbana-Champaign’s Grainger College of Engineering recently published “Online monitoring for safe pedestrian-vehicle interactions." The research paper explores how to “effectively integrate pedestrian intent estimation into an autonomous vehicle stack”, and the possibility of developing an online platform to give “rigorous guarantees on the safety of such human-robot interactions.”
Electrical engineering professors Katherine Driggs-Campbell and Professor Sayan Mitra collaborated on the paper with students Peter Du, Zhe Huang, Tianqi Liu, Tianchen Ji, Ke Xu, Qichao Gao, and Hussein Sibai.
Driggs-Campbell’s research is focused on developing means to create safe autonomous systems which are able to operate out in the real world.
“We were able to demonstrate how our proposed framework was able to successfully and safely maneuver our autonomous vehicle around pedestrians,” Driggs-Campbell said.
She added that one of the key contributions was developing a safety layer that could assess risk or potential collisions (in real-time) and intervene if necessary.
“This work established the feasibility of online safety monitoring and pedestrian intent estimation on a real vehicle platform, but it did so in a relatively narrow set of scenarios,” Mitra said. “The data from these experiments will inform researchers looking to improve the performance and generalizability of these approaches. We also expect the software artifacts being developed for this paper to be picked-up and used in other research projects.”
Mitra researches algorithms for autonomous systems. The group is interested in building the “brains” of an autonomous vehicle ---the core decision and control algorithms that can work reliably in highly dynamic, imperfectly perceived, and uncertain environments.
“Getting these algorithms to work 80% of the time can often be seen as an engineering effort. To achieve 99.9999% reliability---the level of reliability we have come to expect from aircraft control systems---is another story,” he said.
Mitra said the research group is also interested in understanding what the fundamental limits of reliable control are, given the limited computing power and sensing resolution at our disposal.
“We focused a lot of our efforts on building a safe, interactive system from the ground up,” Driggs-Campbell said. “This tight integration of human-robot interaction and formal methods is important for making guarantees in safety critical systems, as we demonstrated in this work.”
From “Online monitoring for safe pedestrian-vehicle interactions”, Mitra said he would like for readers to take away that theoretically sound algorithms for prediction and safety monitoring can indeed be deployed in a real-world autonomous vehicle, and that a small group of Grainger Computer Engineering students were successful in getting the end-to-end system described in the paper up and running in under six months.
“I hope this paper brings to light the importance of taking a human-centered approach to designing autonomy, and demonstrates how critical formal techniques are for safe interactions,” Driggs-Campbell said.