chomeed/mimicgen_square_d0_224x224_dinowm_ph
Updated
World models and task-progress models for online robot policy refinement. Code: https://github.com/chomeed/dvd
Note World model f for square_d0. DVD runs load ph_conv_150k_cont150k (DICE-RL and DSRL); V pretraining used ph_conv_150k. Trained on the prior's rollouts mimicgen_square_d0_224x224_mtdit_flow_40k_{success,failure}, episodes 0-79.
Note Task-progress model (RoboMeter LoRA) for square_d0, used to initialize the value model V.
Note Task-progress model (RoboMeter LoRA) for threading_d0, used to initialize the value model V.
Note Task-progress model (RoboMeter LoRA) for hammer_cleanup_d1, used to initialize the value model V.