Joydeep Biswas, assistant professor in the College of Computer and Information Sciences, has recently been awarded an Amazon Research Award of $80,000 to fund his research on modeling dynamic human environments for robot perception.
“Autonomous mobile robots need to accurately and robustly reason about their location in the environment, along with the locations of other movable objects, to perform tasks accurately," said Biswas. “However, localization in real human environments over extended periods of time is challenging, since the environment changes often. Such changes include changes to the geometry due to movable and deformable objects and changes in appearance due to lighting and weather changes.”
According to Biswas, the key insight underlying his solution is the realization that the objects responsible for changes in the environment are often also the same objects that mobile robots need to detect and track to effectively execute their tasks. His work aims to jointly model and track the many distinct movable objects that compose the world, while simultaneously reasoning about the location of the robot.
His approach has the potential to solve several robot perception problems in a single, principled manner, naturally fusing together a number of perception problems that are conventionally tackled independently, including robot localization and mapping.
“The potential impact of this research is varied,” said Biswas. “It will benefit not only service mobile robots deployed in everyday human environments but it also has the potential to help autonomous cars navigate in a changing world with reduced reliance on up-to-date maps.”