Thomas Lips

allow for anonymous wandb accounts (i.e. no login of the user), this will allow for interacting with the codebase w/o requiring a wandb account (useful for e.g. CI) see https://docs.wandb.ai/ref/app/features/anon

This would catch bugs such as #14. Use the cloth dataset to avoid duplicate work. make sure it also has - multiple instances in a single image - keypoints that...

split dataset into train and val, and log val loss during training

12

## What this does Modifies train script to split dataset into train and val datasets and logs validation loss during training. Based on #158 closes #250 ## How it was...

Question - how to handle additional sensory input

4

Hi guys, sorry to bother you again :wink: and thanks for your work, I'm very excited by Lerobot! I'm currently collecting some teleop data where the robot has tactile sensors...

[Question] How to set training to a local dataset?

2

Is there a way to train on a local dataset without manually adding the `local_files_only` arg to the `make_dataset` function of the train script? I have set the `LEROBOT_HOME` env...

Gello dataset converter

6

I made a converter for the [Gello](https://wuphilipp.github.io/gello_site/) dataset format (pickles containing dicts with all the observations). If this is of interest, I am willing to contribute it back here. The...

The video encoding in the dataset uses ffmpeg through a subprocess call. This imposes a constraint on the file paths: they cannot contain a space as this is considered a...

In the [data collection script](https://github.com/huggingface/lerobot/blob/eec6796cb8f75a87753fb054ffb277d3a28a10e3/lerobot/scripts/control_sim_robot.py#L278) for sim envs, the action (a_t) is determined based on the current observation (o_t). However, when the `step` method is called, the observation is overwritten...

When running the `compute_norm_stats.py` script on a device with multiple GPUs, you get a sharding error because the batch (with size=1) cannot be split across the devices. You can fix...