Sign inGet started
Dataset · Hugging Face

RLE-Bench/booster-t2-integration-acceptance

Booster T2 demo Household task videos with Kinex and GPT-6 Astra: 12 realtime tasks and one separate historical stepped attempt. Each attempt includes its instruction, simulator outcome, token usage and public agent session. The overview videos are original recordings. Head and hand videos were rendered by replaying the original commands in the original simulator environments; they are additional views of those attempts, not new model runs. Replay checks and media provenance… See the full description on the dataset page: https://huggingface.co/datasets/RLE-Bench/booster-t2-integration-acceptan

Author / organizationRLE-Bench
Downloads183 (as reported by Hugging Face)
Likes0
Created21 October 2026
Last updated at source22 October 2026
Embed this entry
<iframe src="https://silytics.si/embed/item/4dfacacc-0377-44b2-9cdd-e59120849c64" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Recent changes

  • 22 Oct 2026 Description: Booster T2 integration acceptance Single-seed Kinex trials of RLE-Bench task11, with RoboEnv Isaac Lab devices. The default protocol is realtime, seed 0, GPT-6 Astra / high, a 3,600-second wall-clock deadline, recording and native stop-on-s → Booster T2 demo Household task videos with Kinex and GPT-6 Astra: 12 realtime tasks and one separate historical stepped attempt. Each attempt includes its instruction, simulator outcome, token usage and public agent session. The overview vi

Changes detected by Silytics between refreshes of the source.

More from RLE-Bench

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 22 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.