metadata
license: apache-2.0
The fine-tuned VLMs and CLIP model used in this work are available at:
https://huggingface.co/ys-qu/found-rl_vlms
The implementation code is available at:
https://github.com/ys-qu/found-rl
We also provide trained RL policy checkpoints for DrQv2-CLIP to facilitate direct evaluation:
drqv2-clip-lb.zip: a checkpoint from one of the three random-seed training runs on the Leaderboard benchmark, evaluated on the Leaderboard benchmark.drqv2-clip-eu.zip: a checkpoint from one of the three random-seed training runs on the NoCrash benchmark, evaluated under the NoCrash setting.
These checkpoints can be used to run evaluation without retraining the RL agents from scratch.