Fix missing type variable in policy confg.json file after training

[anredlich]

What this does

Fixes: when using train.py or, for example, 3_train_policy.py on either a real robot database or a simulated database, the output policy config.json file is missing the policy type line, such as "type": "act". When an evaluation script, such as eval.py or 2_evaluate_pretrained_policy.py tries to read the config.json file it fails with an error:
ParsingError(f"Expected a dict with a '{CHOICE_TYPE_KEY}' key for {cls}, got {raw_value}").

Label: (🐛 Bug)

To fix:
In lerobot/common/policies/act/configuration_act.py:

• Add type: str = "act" (line 94) to class ACTConfig(PreTrainedConfig):

In lerobot/common/policies/diffusion/configuration_diffusion.py:

• Add type: str = "diffusion" (line 107) to class DiffusionConfig(PreTrainedConfig):

How it was tested

Before fix: Evaluating a learned policy, e.g. using eval.py gave ParsingError(f"Expected a dict with a '{CHOICE_TYPE_KEY}' key for {cls}, got {raw_value}") and exited.

After fix: No errors when evaluating, e.g. using eval.py.

Tested scenarios:

• Training a policy from a real robot dataset using train.py then evaluating using eval.py.
• Training a policy from a simulated dataset using, e.g. 3_train_policy.py and then evaluating using 2_evaluate_pretrained_policy.py.

How to checkout & try? (for the reviewer)

Test with the pushT example scripts:

python examples/3_train_policy.py
python examples/2_evaluate_pretrained_policy.py

# Will fail with ParsingError:

ParsingError(f"Expected a dict with a '{CHOICE_TYPE_KEY}' key for {cls}, got {raw_value}")
and config.json in "outputs/train/example_pusht_diffusion" will be missing 
"type": "diffusion"

Check that config.json in "outputs/train/example_pusht_diffusion" contains "type": "diffusion".
Without fixes they will not. With fixes, they will.