metadata

license: apache-2.0
datasets:
  - lerobot/pusht_keypoints
base_model:
  - lerobot/diffusion_pusht_keypoints

Diffusion PushT-v0 using Keypoints

This repository contains the latest checkpoint of the training visible at: https://wandb.ai/fiatlux/diffusion-pusht-keypoints/workspace?nw=nwuserandrearitossa

I am researching for more efficient ways of training diffusion and therefore I am experimenting with the architecture. As a result to replicate or use the model use this branch of "huggingface/lerobot": https://github.com/the-future-dev/lerobot/tree/cloth-diff

Demo Video

Here’s a sample output from the model:

Evaluation

The model was evaluated on the PushT environment from gym-pusht. There are two evaluation metrics on a per-episode basis:

Maximum overlap with target (seen as eval/avg_max_reward in the charts above). This ranges in [0, 1].
Success: whether or not the maximum overlap is at least 95%.

Here are the metrics for 500 episodes worth of evaluation.

Metric	Average over 500 episodes
Average max. overlap ratio	0.9780
Success rate (%)	86.80%

The results of each of the individual rollouts may be found in eval_results.json.