The ability to track, predict and reason about pedestrians and vehicles in a fast-paced dense urban environment is crucial to ensuring that autonomous vehicles can operate safely and dependably. This project focuses on providing that capability to fleets of autonomous cars and delivery drones, allowing these autonomous systems to realize their promised societal benefits, such as the potential for greater mobility of people and goods while reducing traffic congestion and increasing safety. This technology can moreover be customized for other applications such as large manufacturing operations and even small household robotic applications.
The project approaches this challenge in several ways: (i) combine classification algorithms from machine vision with the motion tracking of the objects from multi-target Bayesian filters into a new filtering architecture; (ii) generate new online, distributed tessellation algorithms, using tools from Voronoi-based distributed coverage control, to dynamically partition the surrounding environment in a way that leverages the innate parallelism of teams of multi-robots; (iii) use this partition to create a distributed memory architecture that has bandwidth-efficient updates and ensures data integrity in the face of system errors and malicious agents; (iv) develop semantically-aware path planning algorithms for fast, online optimization of robot motion that account for the range of possible reactionary behaviors of other objects; and (v) design an app-based interface that facilitates two-way information exchange between a human operator and the multi-robot team. A prototype system based on these advances will be tested and evaluated in a highly instrumented laboratory environment.
Robot-Assisted Pedestrian Flow Regulation and Human-Robot Interaction (2016-2017) NSF Award 1527016 [Link]
Having well-designed robotic systems to assist people in public crowd environments such as shopping malls, museums, and campus buildings benefits society economically. More important, in life-threatening emergency situations, robot-assisted evacuation could save lives by reducing congestion and preventing crowd stampede. We develop real time re-configurable crowd control by interacting robots, which replaces costly infrastructure modification for local crowd regulation. Machine learning based methods are used to learn human-robot interaction for effective robot navigation and pedestrian guidance in humans’ environments.