Add New 3 Joint Ragdoll Worm Environment

[Hunter-Unity]

Proposed change(s)

New 3 Joint Ragdoll Worm Environment

Useful links (Github issues, JIRA tickets, ML-Agents forum threads etc.)

Types of change(s)

• Bug fix
• New feature
• Code refactor
• Breaking change
• Documentation update
• Other (please describe)

Checklist

• Added tests that prove my fix is effective or that my feature works
• Updated the changelog (if applicable)
• Updated the documentation (if applicable)
• Updated the migration guide (if applicable)

Other comments

worm are cool i guess

[CLAassistant]

All committers have signed the CLA.

[andrewcoh]

Just curious this changed in the physics manager?

[Hunter-Unity]

I'm not sure what this change to the physics manager is. i didn't intentionally change any of the physics settings that i know of.

[awjuliani]

Overall looking good. Just have a few comments:

• We should only have one WormDynamic.nn file. I think we can delete the 28 million time-step one.
• You'll want to add this environment (pic and description) to the Learning-Environment-Examples.md file in the docs.
• Not for this PR, but once we have DynamicWalker, I am wondering how much of the code can be shared between the three locomotion agent types.
• Once this PR is merged, we will want this added to the daily training runs so we can benchmark it over time.
• Looks like you need to sign the CLA before we can accept the PR.

[Hunter-Unity]

all changes made @awjuliani. i think it's ready for merge

[awjuliani]

This doesn't seem right, since for Crawler, the static reward is 5x the dynamic reward.

[Hunter-Unity]

i retrained to confirm benchmark. it scored 200 so i updated the doc. several reasons there is a huge difference.

1. I increased the gravity to 1.5x so the worm is experiencing more gravity. The crawler can go much faster because it's essentially floating
2. The worm has much weaker joints than the crawler. If the joints are strong it doesn't have natural oscillating motion. weaker joints mean it can't rely on launching itself foward like crawler does.
3. all body segments on the worm are of mass 10. this is to give it a heavier look.

overall these changes were made to give the worm a less floaty look. I expected the benchmark to be lower.

[awjuliani]

Looks like you updated worm dynamic to be higher than worm static. Is that right? I know the overall will be lower for both, but shouldn't static be better than dynamic?

[Hunter-Unity]

yes. error on my part. i updated it.

[awjuliani]

🐍

[chriselion]

Is this going to get merged for the release?