Wheel odometry is the default odometry source for the robot’s base. Based on our experience, the wheel odometry is completely unusable if the robot operates on carpet, which is common indoors.
To overcome this problem, we disabled the wheel odometry entirely and replaced it with 2d laser odometry. It works very well in most situations but there can be instances where the robot gets lost due to bad odometry. There are other alternatives, for instance we could use visual odometry or fuse different sensor inputs to produce a better odometry source.
I wonder what the community of Stretch users uses to replace the wheel odometry of the robot and if some people would like to share their experience. Thank you
Hey Benoit, we have faced this issue before. Our solution was to use “likelihood_field_prob” in laser_model_type instead of “likelihood_field”.
1 Like
@olaghattas Thank you, I will try this parameter of AMCL. But did you keep using the wheel odometry ?
yeah, we kept using the wheel odometry.