Path integration: allow different contact force thresholds for different legs, filter contact force signal

[sibocw]

Description

1. Allow different contact force thresholds (used for stance/swing detection) for fore/mid/hind legs.
2. Filter contact force signal — hindleg contact might be noisy
3. Try path integration over complex terrain unrealistic — arena too big and requires very stable physics.

Types of changes

• Bugfix
• New feature
• Refactor / Code style update (no logical changes)
• Build / CI changes
• Documentation Update
• Other (explain)

Does this address any currently open issues?

[list open issues here]

[sibocw]

Main changes:

• Allow different force thresholds for fore/mid/hind legs
• Added support for blocks terrain, although unused in the end
• Converted build_models.ipynb into a script
• Changed parallelism for training data collection: before, exploration.py was has a CLI and a separate script makes multiple calls to the CLI in parallel. Now, we just use joblib.
• Removed unused legacy code for closed loop homing (not implemented in the end)

Can either @tkclam or @stimpfli take a look please? Thanks!

[sibocw]

(Commits up to here are used for v1_SW of the Figure 4a-c)

[tkclam]

@sibocw The code was blocking indefinitely in pool.imap when I ran it, but works well when I use joblib. Everything else is fine other than that.

[sibocw]

@sibocw The code was blocking indefinitely in pool.imap when I ran it, but works well when I use joblib. Everything else is fine other than that.

Did you have to change the code in the end? any idea why?

[tkclam]

I temporarily changed to using joblib to get it to run. Not sure why multiprocessing didn't work

[sibocw]

do you want to push your version to dev-v1.0.0?

[tkclam]

done