Mobile robot navigation is typically regarded as a geometric problem, in which the robot’s objective is to perceive the geometry of the environment in order to plan collision-free paths towards a desired goal.
However, a purely geometric view of the world can can be insufficient for many navigation problems.
For example, a robot navigating based on geometry may get trapped in a field of tall grass because it believes it is untraversable.
We investigate how to move beyond these purely geometric-based approaches using a method that learns about physical navigational affordances from experience.
Berkeley Autonomous Driving Ground Robot (BADGR)
BADGR is an end-to-end learning-based mobile robot navigation system that can be trained with self-supervised off-policy data gathered in real-world environments, without any simulation or human supervision.
BADGR learns to navigate by:
1. Autonomously collecting data
2. Autonomously labeling the data using self-supervision
3. Training an image-based neural network predictive model
Using this model, BADGR can accurately predict which action sequences will lead to which outcomes, such as bumpiness (left) or collision (right)
BADGR is then able to
navigate in urban environments
navigate in off-road environments
improve as it gathers more data
generalize to never-before-seen environments