Robot Manipulation Data for Reinforcement Learning Finetuning

This dataset supports the DICE-RL framework for refining robot policies via reinforcement learning. It contains pretraining data for behavior cloning and finetuning data across various Robomimic environments. The dataset was created by author 'wintermelontree' and was last updated in March 2026.

Use Cases

• Apply the DICE-RL framework to finetune robot policies using the provided finetuning data.
• Train behavior cloning models on the pretraining data for robot manipulation tasks.
• Benchmark reinforcement learning algorithms on Robomimic environments using the included data.

Strengths

• Data is associated with a published research framework (DICE-RL) for robot policy refinement.
• Includes both pretraining and finetuning data for different stages of robot learning.
• Dataset was updated in March 2026, indicating recent maintenance.

Limitations

• The dataset structure, column names, sample data, and size are unknown from the provided input.
• Specific file formats and license information are not provided.
• The number of rows, columns, and specific task details are unavailable.

Provenance

Source

wintermelontree on Hugging Face

Collection Method

Created for the DICE-RL research project on robot policy finetuning.

Freshness

Last updated on 2026-03-20.