DEP new rules rethink:
With extra information on the rules With better smoothing on the robot sensors/motors Adding more explainations in relation to dynamics
git add .
git commit -m "adding changeable terrain map for exploration and adaption in the complex terrain"
git push -u origin main
- using the learning rules from the paper
DEP rule: https://www.pnas.org/content/112/45/E6224
Supplementary materials: playfulmachines.com/DEP (Ralf Der et al)
- using the diamond structure introduced here:
DIAMOND rule: https://www.frontiersin.org/articles/10.3389/fnbot.2020.00062/full (Simon et al)
- Simulator:
Lpzrobot software: https://github.com/georgmartius/lpzrobots (Georg Martius et al)
Installation Scripts: https://github.com/larsgroeber/lpzrobots-install-script
| uC when Lambda_update_interval = 1 | Lambda_update_interval = 2 |
|---|---|
 |
 |
| Lambda_update_interval = 3 | Lambda_update_interval = 4 |
|---|---|
 |
 |
| Lambda_update_interval = 5 | Lambda_update_interval = 10 |
|---|---|
 |
 |
We see that when Lambda_update_interval=1 and synboost=1.0, the uC (updateC) matrix is an unit matrix, then as the Lambda_update_interval goes up, the uC matrix goes far away from an unit matrix, but still from 2-4, we can also see sometimes it perserves an unit matrix.