Assistant Professor Sylvia Herbert and Associate Professor Sicun (Sean) Gao, alongside Ph.D. students Hongzhan Yu, Chiaki Hirayama, and Chenning Yu, have been honored with the Robocup Best Paper Award at the 2023 IEEE International Conference on Intelligent Robots and Systems (IROS). This achievement stands out among the approximately 1,800 papers presented at the conference, making it a remarkable accomplishment for the team.
Their award-winning paper, titled “Sequential Neural Barriers for Scalable Dynamic Obstacle Avoidance” focuses on enhancing safe robot navigation within crowds of varying densities. Their innovative approach involves training a neural network-based safety function, known as a control barrier function (CBF), to analyze safety in the presence of complex dynamics, such as pedestrian interactions. By decomposing the spatial interaction patterns of dynamic obstacles over time, their method allows safety functions and control policies initially trained with a small number of pedestrians to be effectively generalized to environments with significantly higher pedestrian densities, resulting in improved performance and dynamic collision avoidance when compared to existing methods like potential fields, end-to-end reinforcement learning, and model-predictive control.
Review their detailed Figure here.