This helps make sure that we disable the critics that can create local-minima near goals are disabled closer to the goal poses. Rather than setting it to the positional tolerance (which could be 0.1-0.3m), we decouple when we stop considering these critics from the actual positional tolerance itself so a larger window around goals can be incentivized to focus on the relevant critics (collision, goal, goal angle).
I found that this (and tuning some of the weights a little more, which I later realized were off because my files were outdated :laughing: ) to increase the convergence to goal to nearly every time and doing so in a more direct way with less confusion.
Even after these changes #79 needs to stay open until we can do testing on hardware to make sure that these changes and the previous work by @artofnothingness is sufficient.
SteveMacenski
commented
1 year ago
In further response to #79
This helps make sure that we disable the critics that can create local-minima near goals are disabled closer to the goal poses. Rather than setting it to the positional tolerance (which could be 0.1-0.3m), we decouple when we stop considering these critics from the actual positional tolerance itself so a larger window around goals can be incentivized to focus on the relevant critics (collision, goal, goal angle).
I found that this (and tuning some of the weights a little more, which I later realized were off because my files were outdated :laughing: ) to increase the convergence to goal to nearly every time and doing so in a more direct way with less confusion.
Even after these changes #79 needs to stay open until we can do testing on hardware to make sure that these changes and the previous work by @artofnothingness is sufficient.