Empowering Mobile Manipulators with Internet-of-Things Sensing for Real-world Unstructured Environments

Maintaining a world model is important for
planning mobile manipulation tasks. The process of maintaining
the correspondence between objects in the world model
(symbols) and perceptual data relating to the same physical
objects is called perceptual anchoring. Many approaches
perform anchoring using robot-mounted sensors, while others
instrument the environment with stationary sensors to exploit
the mobility of the robot and the high visual coverage of the
stationary sensors. However, many of the latter works use
custom sensors or make assumptions which are not applicable
in real-world scenarios. We propose that the way towards
world modeling in real-world scenarios is by using off-the-shelf
commercially available sensors. In this paper, we compare this
approach to existing world modeling works and discuss the
challenges that lie ahead.