Jethro's Braindump

Uncertainty in Robotics

Robotic applications increasingly deal with more unstructured environments. Robots that can perceive and deal with uncertainty are much more robust in these scenarios.

Uncertainty arises from:

Environment
unpredictability of the physical world
Sensors
limitations in sensor perception
Robots
robot actuations involve motors that have control noise
Models
models of the world inherently inaccurate
Computation
many algorithms are approximate

A robot that carries a notion of its own uncertainty that acts accordingly is superior to one that does not.

Thrun, Burgard, and Fox, n.d.

Pros

  1. Robust, and scale better to complex, unstructured environments
  2. Weaker requirements on the accuracy of the models compared to classical planning algorithms
  3. Sound methodology for many flavours of robot learning
  4. Broadly applicable to many problems, including perception and action

Cons

  • Relatively computationally inefficient
  • Requires approximation (exact posteriors are computationally intractable)

Links to this note